Upload folder using huggingface_hub

.gitignore

@@ -0,0 +1,2 @@
+__pycache__
+.DS_Store

.gitmodules

@@ -0,0 +1,3 @@
+[submodule "pipeline/utils/uroman"]
+	path = pipeline/utils/uroman
+	url = https://github.com/isi-nlp/uroman.git

LICENSE

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

README.md

@@ -0,0 +1,118 @@
+# GigaSpeech 2
+[![arXiv](https://img.shields.io/badge/arXiv-Paper-COLOR.svg)](https://arxiv.org/pdf/2406.11546)  [![hf](https://img.shields.io/badge/%F0%9F%A4%97%20HuggingFace-Dataset-yellow)](https://huggingface.co/datasets/speechcolab/gigaspeech2)  [![GitHub](https://img.shields.io/badge/GitHub-Repo-green)](https://github.com/SpeechColab/GigaSpeech2)  [![demo](https://img.shields.io/badge/WebPage-Demo-red)](https://huggingface.co/spaces/k2-fsa/automatic-speech-recognition)
+
+This is the official repository of the GigaSpeech 2 dataset. For details of how we created the dataset, please refer to our [arXiv preprint paper](https://arxiv.org/pdf/2406.11546).
+
+GigaSpeech 2 version: 2.0 (2024/06/19)
+
+<div align="left">
+    <p><img src="https://github.com/yfyeung/GigaSpeech2/blob/main/docs/source/_static/pipeline.png" width=800></p>
+</div>
+
+## Download
+* The dataset is available at [HuggingFace](https://huggingface.co/datasets/speechcolab/gigaspeech2) and [ModelScope](https://modelscope.cn/datasets/AI-ModelScope/gigaspeech2).
+* The pre-trained models are available at [Thai](https://huggingface.co/yfyeung/icefall-asr-gigaspeech2-th-zipformer-2024-06-20) and [Vietnamese](https://huggingface.co/zzasdf/icefall-asr-gigaspeech2-vi-zipformer).
+
+## Leaderboard
+
+| **Contributor**| **Toolkit**       | **Train Recipe**     | **Train Data** | **Inference**     |**Test CER/WER**    |
+|:---------------|:------------------|:------------------|:------------------|:------------------|:------------------:|
+|||||
+| <em>Baseline</em>   | [Icefall](https://github.com/k2-fsa/icefall) | Zipformer/Stateless pruned RNN-T | GigaSpeech 2.0 th | TODO | 12.46 | 
+| <em>Baseline</em>   | [Icefall](https://github.com/k2-fsa/icefall) | Zipformer/Stateless pruned RNN-T | GigaSpeech 2.0 id | TODO | 14.92 | 
+| <em>Baseline</em>   | [Icefall](https://github.com/k2-fsa/icefall) | Zipformer/Stateless pruned RNN-T | GigaSpeech 2.0 vi | TODO | 12.83 | 
+| <em>Baseline</em>    | [ESPNet](https://github.com/espnet/espnet) | Conformer/Transformer CTC/AED | GigaSpeech 2.0 th | TODO | 13.70 |
+| <em>Baseline</em>    | [ESPNet](https://github.com/espnet/espnet) | Conformer/Transformer CTC/AED | GigaSpeech 2.0 id | TODO | 15.50 |
+| <em>Baseline</em>    | [ESPNet](https://github.com/espnet/espnet) | Conformer/Transformer CTC/AED | GigaSpeech 2.0 vi | TODO | 14.60 |
+
+## Dataset
+
+### Audio Source
+* Language: Thai, Indonesian, Vietnamese
+* GigaSpeech 2 raw: 30,000 hours of automatically transcribed speech across Thai, Indonesian, and Vietnamese.
+* GigaSpeech 2 refined: 10,000 hours of Thai, 6,000 hours each for Indonesian and Vietnamese.
+* GigaSpeech 2 DEV & TEST: 10 hours for DEV and 10 hours for TEST per language, **transcribed by professional human annotators**, challenging and realistic.
+
+### Training Subsets
+|                      | Thai (hours) | Indonesian (hours) | Vietnamese (hours) |
+|:--------------------:|:------------:|:------------------:|:------------------:|
+| GigaSpeech 2 raw     |    12901.8   |      8112.9        |      7324.0        |
+| GigaSpeech 2 refined |    10262.0   |      5714.0        |      6039.0        |
+
+GigaSpeech 2 raw contains all the data from GigaSpeech 2 refined.
+
+### Evaluation Subsets
+|                      | Thai (hours) | Indonesian (hours) | Vietnamese (hours) |
+|:--------------------:|:------------:|:------------------:|:------------------:|
+| GigaSpeech 2 DEV     |     10.0     |       10.0         |       10.2         |
+| GigaSpeech 2 TEST    |     10.0     |       10.0         |       11.0         |
+
+Evaluation subsets are **annotated by professional human annotators**.
+
+### Preparation Scripts
+Soon available at [Lhotse](https://github.com/lhotse-speech/lhotse).
+
+### Metadata Walkthrough
+Soon available.
+
+### Audio Processing
+GigaSpeech 2 audio files are resampled to 16 kHz and converted to single-channel WAV format. For detailed implementation, refer to [pipeline/convert_transcribe/convert_and_transcribe.py](https://github.com/yfyeung/GigaSpeech2/blob/main/pipeline/convert_transcribe/convert_and_transcribe.py#L45).
+
+### Text Pre-Processing
+Transcripts are normalized by applying NFKC, converting all characters to uppercase, removing punctuation, and mapping Arabic numerals to words in the respective languages.
+
+### Text Post-Processing (before scoring)
+We standardize by applying NFKC, converting all characters to uppercase, removing punctuation, and merging consecutive whitespace or removing all whitespace from both hypothesis and reference text before CER/WER scoring to ensure apple-to-apple performance comparisons across different toolkits or commercial services.
+
+We also provide the following code snippet, which is used in all the experiments reported in our paper and leaderboard.
+
+```python
+import string
+import unicodedata
+
+def text_post_processing(text):
+    text = unicodedata.normalize("NFKC", text)  # apply NFKC
+    text = text.upper()  # convert to uppercase
+    text = text.replace("-", " ")  # remove hyphen
+    text = re.sub("[{}]".format(string.punctuation), "", text)  # remove punctuation
+    text = re.sub(r"\s+", "", text).strip()  # remove all whitespace for Thai
+    return text
+```
+
+## Collaboration
+We are a group of volunteers trying to make speech technologies easier to use. We welcome any kind of contributions. Currently, we are exploring the following directions. If you are interested in one of the directions and you think you will be able to help, please contact gigaspeech@speechcolab.org.
+
+* Inference architecture for different pre-trained models
+* Adding diverse audio source
+* Benchmarking speech algorithms/services
+* Building and releasing pre-trained models
+* Supporting more languages
+* Making new datasets with permissive licenses
+
+## Institutional Contributors
+|  Institution | Contribution |
+|:------|:-----|
+| [Shanghai Jiao Tong University](https://www.seiee.sjtu.edu.cn/) | Computing power; Data host; Researchers |
+| [The Chinese University of Hong Kong](https://www.cuhk.edu.hk/chinese/index.html) | Researchers |
+| [Tsinghua University](https://www.ee.tsinghua.edu.cn/en/) | Researchers |
+| [Seasalt AI](https://seasalt.ai/) | Researchers |
+| [Birch AI](https://birch.ai/) | Researchers |
+| [Peng Cheng Laboratory](https://data-starcloud.pcl.ac.cn/) | Researchers; Computing power |
+| [Dataocean AI](https://en.haitianruisheng.com/) | Evaluation data annotation |
+
+## Citation
+Please cite our paper if you find this work useful:
+```
+@article{gigaspeech2,
+  title={GigaSpeech 2: An Evolving, Large-Scale and Multi-domain ASR Corpus for Low-Resource Languages with Automated Crawling, Transcription and Refinement},
+  author={Yifan Yang and Zheshu Song and Jianheng Zhuo and Mingyu Cui and Jinpeng Li and Bo Yang and Yexing Du and Ziyang Ma and Xunying Liu and Ziyuan Wang and Ke Li and Shuai Fan and Kai Yu and Wei-Qiang Zhang and Guoguo Chen and Xie Chen},
+  journal={arXiv preprint arXiv:2406.11546},
+  year={2024},
+}
+```
+
+## Contact
+If you have any concerns, please contact gigaspeech@speechcolab.org.
+
+## Metadata Changelog
+- 2024/06/19 v2.0: Initial release.

pipeline/convert_transcribe/README.md

@@ -0,0 +1,51 @@
+## Installation
+```shell
+conda install ffmpeg
+pip install ffmpeg-python
+```
+
+### Option 1: Standard Whisper
+
+```shell
+pip install git+https://github.com/openai/whisper.git
+```
+
+### Option 2: Faster whisper
+
+```shell
+pip install faster-whisper
+```
+
+## Usage
+Refer to the language codes in the [Whisper repository](https://github.com/openai/whisper/blob/main/whisper/tokenizer.py#L10-L111). 
+
+```shell
+# Standard Whisper
+python convert_and_transcribe.py
+  --lang [whisper language code] \
+  --root-dir [downloaded audio directory] \
+  --save-dir [output directory]
+
+# Faster Whisper
+python convert_and_transcribe.py
+  --lang [whisper language code] \
+  --root-dir [downloaded audio directory] \
+  --save-dir [output directory] \
+  --use-faster-whisper True
+```
+
+For example:
+```shell
+# Standard Whisper
+python convert_and_transcribe.py
+  --lang zh \
+  --root-dir ./download \
+  --save-dir ./output_trans
+
+# Faster Whisper
+python convert_and_transcribe.py
+  --lang zh \
+  --root-dir ./download \
+  --save-dir ./output_trans \
+  --use-faster-whisper True
+```

pipeline/convert_transcribe/convert_and_transcribe.py

@@ -0,0 +1,244 @@
+import argparse
+import os
+import shutil
+import struct
+import subprocess
+from concurrent.futures import ProcessPoolExecutor
+from functools import partial
+
+from tqdm import tqdm
+
+
+def read_wav_info(filename):
+    with open(filename, "rb") as wav_file:
+        a = wav_file.read(28)
+    sr = struct.unpack("i", a[24:28])[0]
+    channel = struct.unpack("h", a[22:24])[0]
+    length = (struct.unpack("i", a[4:8])[0] - 70) / channel / 2 / sr
+    return sr, length
+
+
+def ffmpeg_convert(file_from, file_to):
+    try:
+        subprocess.call(
+            [
+                "ffmpeg",
+                "-loglevel",
+                "warning",
+                "-y",
+                "-i",
+                file_from,
+                "-ac",
+                "1",
+                "-ar",
+                "16000",
+                "-acodec",
+                "pcm_s16le",
+                file_to,
+            ]
+        )
+    except Exception as e:
+        print(file_to, e)
+    return
+
+
+def video2wav(args):
+    video_format = {".mov", ".avi", ".flv", ".ogg", ".mp4", ".mkv", "webm"}
+    if not os.path.exists(args.wav_dir):
+        os.makedirs(args.wav_dir, exist_ok=True)
+
+    # Start multiprocess
+    executor = ProcessPoolExecutor(max_workers=args.workers)
+    print(f"> Using {args.workers} workers!")
+    futures = []
+    files = os.listdir(args.root_dir)
+    files = set([x for x in files if x[-4:] in video_format])
+    print("Total videos to convert: ", len(files))
+
+    for file_name in files:
+        file_raw = os.path.join(args.root_dir, file_name)
+        file_to = os.path.join(
+            args.wav_dir, file_name.split(".")[-2][-11:] + args.format
+        )
+        if os.path.exists(file_to):
+            continue
+        futures.append(executor.submit(partial(ffmpeg_convert, file_raw, file_to)))
+
+    result_list = [future.result() for future in tqdm(futures)]
+    print(len(result_list), "wavs resampled.")
+
+
+def wav2whisper(args):
+    # remove exsit files
+    wav_dir = (
+        os.path.join(args.wav_dir, args.section)
+        if args.section is not None
+        else args.wav_dir
+    )
+    wavs = os.listdir(wav_dir)
+    exist_files = set([x.split(".")[0] for x in os.listdir(args.sub_dir)])
+    error_files = os.path.join(args.list_dir, "error_files.txt")
+    if os.path.exists(error_files):
+        lines = set(open(error_files, "r", encoding="utf-8").readlines())
+        open(error_files, "w", encoding="utf-8").write("".join(list(lines)))
+        lines = set(x.strip().split("/")[-1][:-4] for x in lines)
+        print(
+            "Recognized: ",
+            len(exist_files),
+            "Language mismatched:",
+            len(lines),
+            "Total skip:",
+            len(exist_files | lines),
+        )
+        exist_files |= lines
+    wavs = [x for x in wavs if x.split(".")[0] not in exist_files]
+
+    if args.use_faster_whisper:
+        from faster_whisper import WhisperModel
+
+        model = WhisperModel(args.model_size, device="cpu", compute_type="int8")
+        for wav in tqdm(wavs):
+            audio_path = os.path.join(wav_dir, wav)
+            segments, info = model.transcribe(audio_path, language=args.lang, beam_size=5)
+            if info.language != args.lang:
+                print(f"Expect {args.lang} Detected: {info.language} {audio_path}")
+                with open(error_files, "a") as f:
+                    f.write(
+                        f"Expect {args.lang} Detected: {info.language} {audio_path}\n"
+                    )
+                continue
+            result_path = os.path.join(args.sub_dir, wav.replace("wav", "txt"))
+            with open(result_path, "w", encoding="utf-8") as f:
+                for segment in segments:
+                    f.write(segment.text + "\n")
+    else:
+        import whisper
+        from whisper.utils import get_writer
+
+        def detect_language(audio_path):
+            # load audio and pad/trim it to fit 30 seconds (start from middle part)
+            audio = whisper.load_audio(audio_path)
+            audio = audio[int(len(audio) / 2) :]
+            audio = whisper.pad_or_trim(audio)
+
+            # make log-Mel spectrogram and move to the same device as the model
+            mel = whisper.log_mel_spectrogram(audio, n_mels=n_mels).to(model.device)
+
+            # detect the spoken language
+            _, probs = model.detect_language(mel)
+            return max(probs, key=probs.get)
+
+        model = whisper.load_model(args.model_size)
+        n_mels = 128 if "large" in args.model_size else 80
+        writer = get_writer("txt", args.sub_dir)
+        for wav in tqdm(wavs):
+            audio_path = os.path.join(wav_dir, wav)
+            lang = detect_language(audio_path)
+            if lang != args.lang:
+                print(f"Expect {args.lang} Detected: {lang} {audio_path}")
+                with open(error_files, "a", encoding="utf-8") as f:
+                    f.write(f"Expect {args.lang} Detected: {lang} {audio_path}\n")
+                continue
+            subtitle = model.transcribe(audio_path, language=lang, beam_size=5)
+            writer(subtitle, audio_path)
+
+
+def construct_corpus(args):
+    wav_dir = (
+        os.path.join(args.wav_dir, args.section)
+        if args.section is not None
+        else args.wav_dir
+    )
+    wavs = os.listdir(wav_dir)
+    wavs = set([x.split(".")[0] for x in wavs if x[-3:] == "wav"])
+    txts = os.listdir(args.sub_dir)
+    txts = set([x.split(".")[0] for x in txts if x[-3:] == "txt"])
+    corpus = os.listdir(args.corpus_dir)
+    corpus = set([x.split(".")[0] for x in corpus])
+    move_list = wavs & txts
+
+    for x in move_list:
+        if x in corpus:
+            continue
+        try:
+            read_wav_info(os.path.join(wav_dir, x + ".wav"))
+        except Exception as e:
+            print(x + ".wav", e)
+            continue
+        shutil.move(
+            os.path.join(wav_dir, x + ".wav"), os.path.join(args.corpus_dir, x + ".wav")
+        )
+        shutil.move(
+            os.path.join(args.sub_dir, x + ".txt"),
+            os.path.join(args.corpus_dir, x + ".txt"),
+        )
+    print("Total files in corpus:", len(os.listdir(args.corpus_dir)))
+
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--lang",
+        type=str,
+        help="See ISO 639-1 codes for supported languages (total: 97).",
+    )
+    parser.add_argument(
+        "--format",
+        type=str,
+        default=".wav",
+        help="Set audio format, find other options in ffmpeg documentation.",
+    )
+    parser.add_argument(
+        "--list-dir", type=str, default="./list", help="Path to save channel lists."
+    )
+    parser.add_argument(
+        "--root-dir",
+        type=str,
+        default="./download",
+        help="Dictionary path of downloaded videos.",
+    )
+    parser.add_argument(
+        "--save-dir",
+        type=str,
+        default="./data",
+        help="Dictionary path to save audio files.",
+    )
+    parser.add_argument(
+        "--section", type=str, default=None, help="Section to transribe."
+    )
+    parser.add_argument(
+        "--model-size",
+        type=str,
+        default="large-v3",
+        help="Whisper model size (large, medium, small, base, tiny)",
+    )
+    parser.add_argument(
+        "--use-faster-whisper",
+        type=bool,
+        default=False,
+        help="Whether to use faster-whisper",
+    )
+    parser.add_argument("--workers", type=int, default=16, help="Multiprocess workers.")
+    args = parser.parse_args()
+
+    if not os.path.exists(args.save_dir):
+        os.makedirs(args.save_dir, exist_ok=True)
+    args.sub_dir = os.path.join(args.save_dir, "whisper")
+    args.wav_dir = os.path.join(args.save_dir, "audios")
+    args.corpus_dir = os.path.join(args.save_dir, "corpus")
+    os.makedirs(args.sub_dir, exist_ok=True)
+    os.makedirs(args.wav_dir, exist_ok=True)
+    os.makedirs(args.corpus_dir, exist_ok=True)
+
+    # convert video to wav
+    video2wav(args)
+
+    # process subtitle and video info
+    wav2whisper(args)
+
+    # Pair wav and txt file
+    construct_corpus(args)
+
+
+if __name__ == "__main__":
+    main()

pipeline/crawler/README.md

@@ -0,0 +1,28 @@
+## Installation
+```shell
+pip install git+https://github.com/yt-dlp/yt-dlp.git
+```
+
+## Usage
+### Create a list of youtube channels
+You need to create a list of YouTube channel names from which you intend to download audio. Save this list in a text file (e.g., `zh_channels.txt`) using the format `[channel name]\t@[channel id]`.
+
+For example:
+```
+Youth With You  @iQIYIYouthWithYou
+KUN  @kun_global
+Kun's Official Channel  @kunsofficialchannel6831
+```
+
+### Get the cookies from the browser
+You need to save the cookies in a text file (e.g., `cookies.txt`) using the Netscape format.
+
+### Start the download process
+```shell
+./download_from_youtube_channels.sh [channels list file] [download directory]
+```
+
+For example:
+```shell
+./download_from_youtube_channels.sh zh_channels.txt ./download
+```

pipeline/crawler/cookies.txt

@@ -0,0 +1,12 @@
+# Netscape HTTP Cookie File
+# This file is generated by yt-dlp.  Do not edit.
+
+.youtube.com	TRUE	/	FALSE	0	PREF	hl=en&tz=UTC
+.youtube.com	TRUE	/	TRUE	0	SOCS	CAI
+.youtube.com	TRUE	/	TRUE	1742631660	GPS	1
+.youtube.com	TRUE	/	TRUE	0	YSC	hBRthzI-Dbc
+.youtube.com	TRUE	/	TRUE	1758181861	__Secure-ROLLOUT_TOKEN	CIGJ2PGoofq90QEQ4vPOjpqdjAMYoNz4jpqdjAM%3D
+.youtube.com	TRUE	/	TRUE	1758182110	VISITOR_INFO1_LIVE	cRUSCnYG5HQ
+.youtube.com	TRUE	/	TRUE	1758182110	VISITOR_PRIVACY_METADATA	CgJWThIEGgAgSg%3D%3D
+.youtube.com	TRUE	/	TRUE	1805702110	__Secure-YT_TVFAS	t=484063&s=2
+.youtube.com	TRUE	/	TRUE	1758182110	DEVICE_INFO	ChxOelE0TkRVek9ESTJPREk1TmpnNE9EVTFOUT09EN7Z+b4GGObX+b4G

pipeline/crawler/download_from_youtube_channels.sh

@@ -0,0 +1,23 @@
+#! /usr/bin/bash
+
+echo "Read channels from file: $1, save to: $2"
+
+log_dir=log
+
+if [ ! -d $log_dir ]; then
+  mkdir $log_dir
+fi
+
+while read rows
+do
+  echo "doc duoc:$rows"
+  channel=`echo "$rows" | awk -F"\t" '{print $2}'`
+  channel_name=`echo ${channel:1}`
+  echo "Processing channel: $channel, channel name: $channel_name"
+  yt-dlp -f 'ba' \
+    --download-archive $log_dir/audios_$channel_name.txt \
+    --cookies cookies.txt \
+    https://www.youtube.com/${channel}/videos -o ${2}/$channel_name'/%(title).20s#%(channel_id)s#%(id)s_%(duration)s.%(ext)s' \
+    > $log_dir/$channel_name.log
+done < $1
+echo "Finished."

pipeline/crawler/downloadgs2/xinchaotoilavanvo/Tây Du Ký (Phần 3)#UCbEGdsixsjZJCKYE8rergSw#OgD3IyAzX_E_972.webm

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f198bf0316115cb24d03be935e6effb5464dd88a5beab841779fcdd3ba74fa24
+size 15827494

pipeline/crawler/kenh.txt

@@ -0,0 +1 @@
+Văn Vở	@xinchaotoilavanvo

pipeline/crawler/log/audios_xinchaotoilavanvo.txt

@@ -0,0 +1,24 @@
+youtube 5WOvQG9NaxQ
+youtube h9VS1gdSwzY
+youtube bv-xny0u2fQ
+youtube nJDb-HyMOXo
+youtube B7Pjc4C_ogA
+youtube 9B1YeDReuGE
+youtube iEoTNGXQL2I
+youtube YBf-tAL8EAs
+youtube FG0z-xkHNps
+youtube 5GTaloeY3EQ
+youtube rXfV60LJiZU
+youtube gRR4NrXSWPo
+youtube HTCV5axjk_s
+youtube 2P6dfJldpsE
+youtube egWDv6pRhDY
+youtube 1MTVpK1zQpE
+youtube IvoxwgtWWv8
+youtube -qwS3q_Lu3c
+youtube X9ldeKXTGRg
+youtube P0ezv6GBoSU
+youtube z_vn3fv-EF4
+youtube OgD3IyAzX_E
+youtube 4673DtpQkkU
+youtube zechn7_mUrE

pipeline/crawler/log/xinchaotoilavanvo.log

@@ -0,0 +1,252 @@
+[youtube:tab] Extracting URL: https://www.youtube.com/@xinchaotoilavanvo/videos
+[youtube:tab] @xinchaotoilavanvo/videos: Downloading webpage
+[download] Downloading playlist: Vn V - Videos
+[youtube:tab] UCbEGdsixsjZJCKYE8rergSw page 1: Downloading API JSON
+[youtube:tab] Playlist Vn V - Videos: Downloading 56 items of 56
+[download] Downloading item 1 of 56
+[youtube] Extracting URL: https://www.youtube.com/watch?v=5WOvQG9NaxQ
+[youtube] 5WOvQG9NaxQ: Downloading webpage
+[youtube] 5WOvQG9NaxQ: Downloading tv client config
+[youtube] 5WOvQG9NaxQ: Downloading player 69f581a5
+[youtube] 5WOvQG9NaxQ: Downloading tv player API JSON
+[youtube] 5WOvQG9NaxQ: Downloading ios player API JSON
+[youtube] 5WOvQG9NaxQ: Downloading m3u8 information
+[info] 5WOvQG9NaxQ: Downloading 1 format(s): 251
+[download] Destination: downloadgs2\xinchaotoilavanvo\Nh�n qu theo quan n#UCbEGdsixsjZJCKYE8rergSw#5WOvQG9NaxQ_1156.webm
+
+[download] Downloading item 2 of 56
+[youtube] Extracting URL: https://www.youtube.com/watch?v=h9VS1gdSwzY
+[youtube] h9VS1gdSwzY: Downloading webpage
+[youtube] h9VS1gdSwzY: Downloading tv client config
+[youtube] h9VS1gdSwzY: Downloading tv player API JSON
+[youtube] h9VS1gdSwzY: Downloading ios player API JSON
+[youtube] h9VS1gdSwzY: Downloading m3u8 information
+[info] h9VS1gdSwzY: Downloading 1 format(s): 251
+[download] Destination: downloadgs2\xinchaotoilavanvo\Harry Potter v� ph�n#UCbEGdsixsjZJCKYE8rergSw#h9VS1gdSwzY_2345.webm
+
+[download] Downloading item 3 of 56
+[youtube] Extracting URL: https://www.youtube.com/watch?v=bv-xny0u2fQ
+[youtube] bv-xny0u2fQ: Downloading webpage
+[youtube] bv-xny0u2fQ: Downloading tv client config
+[youtube] bv-xny0u2fQ: Downloading tv player API JSON
+[youtube] bv-xny0u2fQ: Downloading ios player API JSON
+[youtube] bv-xny0u2fQ: Downloading m3u8 information
+[info] bv-xny0u2fQ: Downloading 1 format(s): 251
+[download] Destination: downloadgs2\xinchaotoilavanvo\Thi i anh h�ng tr#UCbEGdsixsjZJCKYE8rergSw#bv-xny0u2fQ_1572.webm
+
+[download] Downloading item 4 of 56
+[youtube] Extracting URL: https://www.youtube.com/watch?v=nJDb-HyMOXo
+[youtube] nJDb-HyMOXo: Downloading webpage
+[youtube] nJDb-HyMOXo: Downloading tv client config
+[youtube] nJDb-HyMOXo: Downloading tv player API JSON
+[youtube] nJDb-HyMOXo: Downloading ios player API JSON
+[youtube] nJDb-HyMOXo: Downloading m3u8 information
+[info] nJDb-HyMOXo: Downloading 1 format(s): 251
+[download] Destination: downloadgs2\xinchaotoilavanvo\Thi i anh h�ng tr#UCbEGdsixsjZJCKYE8rergSw#nJDb-HyMOXo_1813.webm
+
+[download] Downloading item 5 of 56
+[youtube] Extracting URL: https://www.youtube.com/watch?v=B7Pjc4C_ogA
+[youtube] B7Pjc4C_ogA: Downloading webpage
+[youtube] B7Pjc4C_ogA: Downloading tv client config
+[youtube] B7Pjc4C_ogA: Downloading tv player API JSON
+[youtube] B7Pjc4C_ogA: Downloading ios player API JSON
+[youtube] B7Pjc4C_ogA: Downloading m3u8 information
+[info] B7Pjc4C_ogA: Downloading 1 format(s): 251
+[download] Destination: downloadgs2\xinchaotoilavanvo\Nhng c�u chuyn hay#UCbEGdsixsjZJCKYE8rergSw#B7Pjc4C_ogA_2264.webm
+
+[download] Downloading item 6 of 56
+[youtube] Extracting URL: https://www.youtube.com/watch?v=9B1YeDReuGE
+[youtube] 9B1YeDReuGE: Downloading webpage
+[youtube] 9B1YeDReuGE: Downloading tv client config
+[youtube] 9B1YeDReuGE: Downloading tv player API JSON
+[youtube] 9B1YeDReuGE: Downloading ios player API JSON
+[youtube] 9B1YeDReuGE: Downloading m3u8 information
+[info] 9B1YeDReuGE: Downloading 1 format(s): 251
+[download] Destination: downloadgs2\xinchaotoilavanvo\Gii th�ch to�n b v#UCbEGdsixsjZJCKYE8rergSw#9B1YeDReuGE_2932.webm
+
+[download] Downloading item 7 of 56
+[youtube] Extracting URL: https://www.youtube.com/watch?v=iEoTNGXQL2I
+[youtube] iEoTNGXQL2I: Downloading webpage
+[youtube] iEoTNGXQL2I: Downloading tv client config
+[youtube] iEoTNGXQL2I: Downloading tv player API JSON
+[youtube] iEoTNGXQL2I: Downloading ios player API JSON
+[youtube] iEoTNGXQL2I: Downloading m3u8 information
+[info] iEoTNGXQL2I: Downloading 1 format(s): 251
+[download] Destination: downloadgs2\xinchaotoilavanvo\S thi Gilgamesh ph#UCbEGdsixsjZJCKYE8rergSw#iEoTNGXQL2I_1655.webm
+
+[download] Downloading item 8 of 56
+[youtube] Extracting URL: https://www.youtube.com/watch?v=YBf-tAL8EAs
+[youtube] YBf-tAL8EAs: Downloading webpage
+[youtube] YBf-tAL8EAs: Downloading tv client config
+[youtube] YBf-tAL8EAs: Downloading tv player API JSON
+[youtube] YBf-tAL8EAs: Downloading ios player API JSON
+[youtube] YBf-tAL8EAs: Downloading m3u8 information
+[info] YBf-tAL8EAs: Downloading 1 format(s): 251
+[download] Destination: downloadgs2\xinchaotoilavanvo\S thi Gilgamesh ph#UCbEGdsixsjZJCKYE8rergSw#YBf-tAL8EAs_1390.webm
+
+[download] Downloading item 9 of 56
+[youtube] Extracting URL: https://www.youtube.com/watch?v=FG0z-xkHNps
+[youtube] FG0z-xkHNps: Downloading webpage
+[youtube] FG0z-xkHNps: Downloading tv client config
+[youtube] FG0z-xkHNps: Downloading tv player API JSON
+[youtube] FG0z-xkHNps: Downloading ios player API JSON
+[youtube] FG0z-xkHNps: Downloading m3u8 information
+[info] FG0z-xkHNps: Downloading 1 format(s): 251
+[download] Destination: downloadgs2\xinchaotoilavanvo\To�n b v 81 kip n#UCbEGdsixsjZJCKYE8rergSw#FG0z-xkHNps_2348.webm
+
+[download] Downloading item 10 of 56
+[youtube] Extracting URL: https://www.youtube.com/watch?v=5GTaloeY3EQ
+[youtube] 5GTaloeY3EQ: Downloading webpage
+[youtube] 5GTaloeY3EQ: Downloading tv client config
+[youtube] 5GTaloeY3EQ: Downloading tv player API JSON
+[youtube] 5GTaloeY3EQ: Downloading ios player API JSON
+[youtube] 5GTaloeY3EQ: Downloading m3u8 information
+[info] 5GTaloeY3EQ: Downloading 1 format(s): 251
+[download] Destination: downloadgs2\xinchaotoilavanvo\H�n S Tranh H�ng ph#UCbEGdsixsjZJCKYE8rergSw#5GTaloeY3EQ_1538.webm
+
+[download] Downloading item 11 of 56
+[youtube] Extracting URL: https://www.youtube.com/watch?v=rXfV60LJiZU
+[youtube] rXfV60LJiZU: Downloading webpage
+[youtube] rXfV60LJiZU: Downloading tv client config
+[youtube] rXfV60LJiZU: Downloading tv player API JSON
+[youtube] rXfV60LJiZU: Downloading ios player API JSON
+[youtube] rXfV60LJiZU: Downloading m3u8 information
+[info] rXfV60LJiZU: Downloading 1 format(s): 251
+[download] Destination: downloadgs2\xinchaotoilavanvo\H�n S Tranh H�ng ph#UCbEGdsixsjZJCKYE8rergSw#rXfV60LJiZU_1411.webm
+
+[download] Downloading item 12 of 56
+[youtube] Extracting URL: https://www.youtube.com/watch?v=gRR4NrXSWPo
+[youtube] gRR4NrXSWPo: Downloading webpage
+[youtube] gRR4NrXSWPo: Downloading tv client config
+[youtube] gRR4NrXSWPo: Downloading tv player API JSON
+[youtube] gRR4NrXSWPo: Downloading ios player API JSON
+[youtube] gRR4NrXSWPo: Downloading m3u8 information
+[info] gRR4NrXSWPo: Downloading 1 format(s): 251
+[download] Destination: downloadgs2\xinchaotoilavanvo\Xu�n Thu - Chin Qu#UCbEGdsixsjZJCKYE8rergSw#gRR4NrXSWPo_5090.webm
+
+[download] Downloading item 13 of 56
+[youtube] Extracting URL: https://www.youtube.com/watch?v=HTCV5axjk_s
+[youtube] HTCV5axjk_s: Downloading webpage
+[youtube] HTCV5axjk_s: Downloading tv client config
+[youtube] HTCV5axjk_s: Downloading tv player API JSON
+[youtube] HTCV5axjk_s: Downloading ios player API JSON
+[youtube] HTCV5axjk_s: Downloading m3u8 information
+[info] HTCV5axjk_s: Downloading 1 format(s): 251
+[download] Destination: downloadgs2\xinchaotoilavanvo\Thi Chin Quc tron#UCbEGdsixsjZJCKYE8rergSw#HTCV5axjk_s_2593.webm
+
+[download] Downloading item 14 of 56
+[youtube] Extracting URL: https://www.youtube.com/watch?v=2P6dfJldpsE
+[youtube] 2P6dfJldpsE: Downloading webpage
+[youtube] 2P6dfJldpsE: Downloading tv client config
+[youtube] 2P6dfJldpsE: Downloading tv player API JSON
+[youtube] 2P6dfJldpsE: Downloading ios player API JSON
+[youtube] 2P6dfJldpsE: Downloading m3u8 information
+[info] 2P6dfJldpsE: Downloading 1 format(s): 251
+[download] Destination: downloadgs2\xinchaotoilavanvo\Gii th�ch Thn Tho#UCbEGdsixsjZJCKYE8rergSw#2P6dfJldpsE_4086.webm
+
+[download] Downloading item 15 of 56
+[youtube] Extracting URL: https://www.youtube.com/watch?v=egWDv6pRhDY
+[youtube] egWDv6pRhDY: Downloading webpage
+[youtube] egWDv6pRhDY: Downloading tv client config
+[youtube] egWDv6pRhDY: Downloading tv player API JSON
+[youtube] egWDv6pRhDY: Downloading ios player API JSON
+[youtube] egWDv6pRhDY: Downloading m3u8 information
+[info] egWDv6pRhDY: Downloading 1 format(s): 251
+[download] Destination: downloadgs2\xinchaotoilavanvo\Gii th�ch Thi Xu�n#UCbEGdsixsjZJCKYE8rergSw#egWDv6pRhDY_2579.webm
+
+[download] Downloading item 16 of 56
+[youtube] Extracting URL: https://www.youtube.com/watch?v=1MTVpK1zQpE
+[youtube] 1MTVpK1zQpE: Downloading webpage
+[youtube] 1MTVpK1zQpE: Downloading tv client config
+[youtube] 1MTVpK1zQpE: Downloading tv player API JSON
+[youtube] 1MTVpK1zQpE: Downloading ios player API JSON
+[youtube] 1MTVpK1zQpE: Downloading m3u8 information
+[info] 1MTVpK1zQpE: Downloading 1 format(s): 251
+[download] Destination: downloadgs2\xinchaotoilavanvo\Thn Thoi Bc �u (P#UCbEGdsixsjZJCKYE8rergSw#1MTVpK1zQpE_1551.webm
+
+[download] Downloading item 17 of 56
+[youtube] Extracting URL: https://www.youtube.com/watch?v=IvoxwgtWWv8
+[youtube] IvoxwgtWWv8: Downloading webpage
+[youtube] IvoxwgtWWv8: Downloading tv client config
+[youtube] IvoxwgtWWv8: Downloading tv player API JSON
+[youtube] IvoxwgtWWv8: Downloading ios player API JSON
+[youtube] IvoxwgtWWv8: Downloading m3u8 information
+[info] IvoxwgtWWv8: Downloading 1 format(s): 251
+[download] Destination: downloadgs2\xinchaotoilavanvo\Gii th�ch to�n b T#UCbEGdsixsjZJCKYE8rergSw#IvoxwgtWWv8_3079.webm
+
+[download] Downloading item 18 of 56
+[youtube] Extracting URL: https://www.youtube.com/watch?v=-qwS3q_Lu3c
+[youtube] -qwS3q_Lu3c: Downloading webpage
+[youtube] -qwS3q_Lu3c: Downloading tv client config
+[youtube] -qwS3q_Lu3c: Downloading tv player API JSON
+[youtube] -qwS3q_Lu3c: Downloading ios player API JSON
+[youtube] -qwS3q_Lu3c: Downloading m3u8 information
+[info] -qwS3q_Lu3c: Downloading 1 format(s): 251
+[download] Destination: downloadgs2\xinchaotoilavanvo\Harry Potter v� h�n #UCbEGdsixsjZJCKYE8rergSw#-qwS3q_Lu3c_1496.webm
+
+[download] Downloading item 19 of 56
+[youtube] Extracting URL: https://www.youtube.com/watch?v=X9ldeKXTGRg
+[youtube] X9ldeKXTGRg: Downloading webpage
+[youtube] X9ldeKXTGRg: Downloading tv client config
+[youtube] X9ldeKXTGRg: Downloading tv player API JSON
+[youtube] X9ldeKXTGRg: Downloading ios player API JSON
+[youtube] X9ldeKXTGRg: Downloading m3u8 information
+[info] X9ldeKXTGRg: Downloading 1 format(s): 251
+[download] Destination: downloadgs2\xinchaotoilavanvo\Tng quan v thn th#UCbEGdsixsjZJCKYE8rergSw#X9ldeKXTGRg_1317.webm
+
+[download] Downloading item 20 of 56
+[youtube] Extracting URL: https://www.youtube.com/watch?v=P0ezv6GBoSU
+[youtube] P0ezv6GBoSU: Downloading webpage
+[youtube] P0ezv6GBoSU: Downloading tv client config
+[youtube] P0ezv6GBoSU: Downloading tv player API JSON
+[youtube] P0ezv6GBoSU: Downloading ios player API JSON
+[youtube] P0ezv6GBoSU: Downloading m3u8 information
+[info] P0ezv6GBoSU: Downloading 1 format(s): 251
+[download] Destination: downloadgs2\xinchaotoilavanvo\Thn Thoi Bc �u (P#UCbEGdsixsjZJCKYE8rergSw#P0ezv6GBoSU_1141.webm
+
+[download] Downloading item 21 of 56
+[youtube] Extracting URL: https://www.youtube.com/watch?v=z_vn3fv-EF4
+[youtube] z_vn3fv-EF4: Downloading webpage
+[youtube] z_vn3fv-EF4: Downloading tv client config
+[youtube] z_vn3fv-EF4: Downloading tv player API JSON
+[youtube] z_vn3fv-EF4: Downloading ios player API JSON
+[youtube] z_vn3fv-EF4: Downloading m3u8 information
+[info] z_vn3fv-EF4: Downloading 1 format(s): 251
+[download] Destination: downloadgs2\xinchaotoilavanvo\Thn Thoi Bc �u (P#UCbEGdsixsjZJCKYE8rergSw#z_vn3fv-EF4_1329.webm
+
+[download] Downloading item 22 of 56
+[youtube] Extracting URL: https://www.youtube.com/watch?v=OgD3IyAzX_E
+[youtube] OgD3IyAzX_E: Downloading webpage
+[youtube] OgD3IyAzX_E: Downloading tv client config
+[youtube] OgD3IyAzX_E: Downloading tv player API JSON
+[youtube] OgD3IyAzX_E: Downloading ios player API JSON
+[youtube] OgD3IyAzX_E: Downloading m3u8 information
+[info] OgD3IyAzX_E: Downloading 1 format(s): 251
+[download] Destination: downloadgs2\xinchaotoilavanvo\T�y Du K� (Phn 3)#UCbEGdsixsjZJCKYE8rergSw#OgD3IyAzX_E_972.webm
+
+[download] Downloading item 23 of 56
+[youtube] Extracting URL: https://www.youtube.com/watch?v=4673DtpQkkU
+[youtube] 4673DtpQkkU: Downloading webpage
+[youtube] 4673DtpQkkU: Downloading tv client config
+[youtube] 4673DtpQkkU: Downloading tv player API JSON
+[youtube] 4673DtpQkkU: Downloading ios player API JSON
+[youtube] 4673DtpQkkU: Downloading m3u8 information
+[info] 4673DtpQkkU: Downloading 1 format(s): 251
+[download] Destination: downloadgs2\xinchaotoilavanvo\T�y Du K� (Phn 2)#UCbEGdsixsjZJCKYE8rergSw#4673DtpQkkU_1013.webm
+
+[download] Downloading item 24 of 56
+[youtube] Extracting URL: https://www.youtube.com/watch?v=zechn7_mUrE
+[youtube] zechn7_mUrE: Downloading webpage
+[youtube] zechn7_mUrE: Downloading tv client config
+[youtube] zechn7_mUrE: Downloading tv player API JSON
+[youtube] zechn7_mUrE: Downloading ios player API JSON
+[youtube] zechn7_mUrE: Downloading m3u8 information
+[info] zechn7_mUrE: Downloading 1 format(s): 251
+[download] Destination: downloadgs2\xinchaotoilavanvo\To�n b v Thy H t#UCbEGdsixsjZJCKYE8rergSw#zechn7_mUrE_4505.webm
+
+[download] Downloading item 25 of 56
+[youtube] Extracting URL: https://www.youtube.com/watch?v=72Fot34ipYk
+[youtube] 72Fot34ipYk: Downloading webpage
+[youtube] 72Fot34ipYk: Downloading tv client config
+[youtube] 72Fot34ipYk: Downloading tv player API JSON
+[youtube] 72Fot34ipYk: Downloading ios player API JSON

pipeline/force_alignment/README.md

@@ -0,0 +1,18 @@
+## Installation
+```shell
+git submodule update --init --recursive
+pip install torchaudio==2.1.0  # >= 2.1.0
+pip install sox 
+pip install dataclasses 
+```
+
+## Usage
+You need to specify the [ISO 639-2 language code](https://en.wikipedia.org/wiki/List_of_ISO_639_language_codes).
+```shell
+./force_align.sh [transcription directory] [output directory] [ISO 639-2 language code]
+```
+
+For example:
+```shell
+./force_align.sh ./output_trans ./output_force_align zho
+```

pipeline/force_alignment/calculate_precision.py

@@ -0,0 +1,70 @@
+import argparse
+import json
+from pathlib import Path
+
+import textgrid
+
+
+def parse_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--hyp-dir", type=Path, required=True)
+    parser.add_argument("--ref-dir", type=Path, required=True)
+    return parser.parse_args()
+
+
+def calculate_overlap_time(start1, end1, start2, end2):
+    return max(0, min(end1, end2) - max(start1, start2))
+
+
+def calculate_precision(hyp_intervals, ref_intervals):
+    total_true_positive_time = 0
+    total_predicted_time = 0
+
+    for hyp_start, hyp_end in hyp_intervals:
+        hyp_duration = hyp_end - hyp_start
+        total_predicted_time += hyp_duration
+        overlap_time_with_all_refs = 0
+
+        for ref_start, ref_end in ref_intervals:
+            overlap = calculate_overlap_time(hyp_start, hyp_end, ref_start, ref_end)
+            overlap_time_with_all_refs += overlap
+
+        total_true_positive_time += overlap_time_with_all_refs
+
+    precision = total_true_positive_time / total_predicted_time
+
+    return precision
+
+
+def main():
+    args = parse_args()
+
+    for ref_path in args.ref_dir.rglob("*.TextGrid"):
+        file_id = ref_path.stem
+        hyp_path = args.hyp_dir / (file_id + "_manifest.jsonl")
+
+        assert hyp_path.exists(), f"{hyp_path} does not exist."
+
+        tg = textgrid.TextGrid.fromFile(ref_path)
+        ref_intervals = []
+        for interval in tg[0]:
+            if len(interval.mark) > 0:
+                start = interval.minTime
+                end = interval.maxTime
+                ref_intervals.append((start, end))
+
+        hyp_intervals = []
+        with open(hyp_path, "r") as f:
+            for line in f:
+                data = json.loads(line)
+                if len(data["text"]) > 0:
+                    start = data["audio_start_sec"]
+                    duration = data["duration"]
+                    end = start + duration
+                    hyp_intervals.append((start, end))
+
+        print(file_id, calculate_precision(hyp_intervals, ref_intervals))
+
+
+if __name__ == "__main__":
+    main()

pipeline/force_alignment/force_align.sh

@@ -0,0 +1,18 @@
+#! /usr/bin/bash
+
+corpus_dir=${1%/}
+output_dir=${2%/}
+lang=$3
+
+for text in ${corpus_dir}/*.txt; do
+  id=$(basename "${text}" .txt)
+  echo "Process $id"
+  wav="${corpus_dir}/${id}.wav"
+
+  python /content/drive/MyDrive/GigaSpeech2/pipeline/utils/force_alignment/align.py \
+    -a $wav \
+    -t $text \
+    --lang $lang \
+    --output-dir $output_dir \
+    --uroman /content/drive/MyDrive/GigaSpeech2/pipeline/utils/uroman/bin
+done

pipeline/force_alignment/force_align_from_list.sh

@@ -0,0 +1,18 @@
+#! /usr/bin/bash
+
+corpus_dir=${2%/}
+output_dir=${3%/}
+lang=$4
+
+while IFS= read -r text; do
+  id=$(basename "${text}" .txt)
+  echo "Process $id"
+  wav="${corpus_dir}/${id}.wav"
+
+  python ../utils/force_alignment/align.py \
+    -a $wav \
+    -t $text \
+    --lang $lang \
+    --output-dir $output_dir \
+    --uroman ../utils/uroman/bin
+done < $1

pipeline/segmentation/README.md

@@ -0,0 +1,41 @@
+## Usage
+### Installation
+```shell
+pip install fasttext
+```
+
+### Download the language identification model
+```shell
+wget -P /tmp https://dl.fbaipublicfiles.com/fasttext/supervised-models/lid.176.bin
+```
+
+### Modify the filtering rules
+```shell
+vim filter_manifest.py
+```
+
+### Filter the segments
+```shell
+python filter_manifest.py \
+  --input-dir [forced-aligned manifests directory] \
+  --output-dir [output directory] \
+  --lid-model-path [path to lid.176.bin]
+```
+
+For example:
+```shell
+python filter_manifest.py \
+  --input-dir ./output_force_align \
+  --output-dir ./output_filter \
+  --lid-model-path ./lid.176.bin
+```
+
+### Segmentation the audio files
+```shell
+./segment.sh [filtered manifests directory] [output directory]
+```
+
+For example:
+```shell
+./segment.sh ./output_filter ./output_segment
+```

pipeline/segmentation/filter_manifest.py

@@ -0,0 +1,103 @@
+import argparse
+import json
+import re
+from abc import ABC, abstractmethod
+from pathlib import Path
+
+import fasttext
+from tqdm import tqdm
+
+
+class FilterStrategy(ABC):
+    @abstractmethod
+    def apply(self, line):
+        pass
+
+
+class CharsetFilter(FilterStrategy):
+    def __init__(self):
+        thai_chars = r"\u0E00-\u0E7F"
+        digits = r"\u0030-\u0039"
+        blank_symbol = r"\s"
+        valid_symbols = thai_chars + digits + blank_symbol
+        self.valid_pattern = re.compile(f"[^{valid_symbols}]")
+
+    def apply(self, line):
+        return not self.valid_pattern.search(line["text"])
+
+
+class LanguageConfidenceFilter(FilterStrategy):
+    def __init__(self, model_path, confidence_threshold=0.95):
+        self.model = fasttext.load_model(model_path)
+        self.confidence_threshold = confidence_threshold
+
+    def apply(self, line):
+        labels, probabilities = self.model.predict(line["text"], k=1)
+        return probabilities[0] >= self.confidence_threshold
+
+
+class AudioDurationFilter(FilterStrategy):
+    def __init__(self, min_keep_duration=1, max_keep_duration=30):
+        self.min_keep_duration = min_keep_duration
+        self.max_keep_duration = max_keep_duration
+
+    def apply(self, line):
+        return (
+            line["duration"] >= self.min_keep_duration
+            and line["duration"] <= self.max_keep_duration
+        )
+
+
+class ContentFilter:
+    def __init__(self, strategies):
+        self.strategies = strategies
+
+    def __call__(self, line):
+        for strategy in self.strategies:
+            if not strategy.apply(line):
+                return False
+        return True
+
+
+def parse_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--input-dir", type=Path, required=True)
+    parser.add_argument("--output-dir", type=Path, required=True)
+    parser.add_argument("--lid-model-path", type=str, required=True)
+    return parser.parse_args()
+
+
+def filter_manifests(input_dir, output_dir, content_filter):
+    total_cnt = 0
+    valid_cnt = 0
+    for manifest_path in tqdm(input_dir.rglob("*.jsonl"), desc="Filtering manifests"):
+        filtered_manifest_path = output_dir / ("filtered_" + manifest_path.name)
+
+        with open(manifest_path, "r", encoding="utf-8") as reader, open(
+            filtered_manifest_path, "w", encoding="utf-8"
+        ) as writer:
+            for line in reader:
+                line = json.loads(line)
+                total_cnt += 1
+                if content_filter(line):
+                    writer.write(json.dumps(line) + "\n")
+                    valid_cnt += 1
+
+    print(
+        f"total segments: {total_cnt}, valid segments: {valid_cnt}, filtered rate: {1 - valid_cnt / total_cnt}"
+    )
+
+
+def main():
+    args = parse_args()
+    strategies = [
+        CharsetFilter(),
+        LanguageConfidenceFilter(args.lid_model_path, 0.99),
+        AudioDurationFilter(2, 30),
+    ]
+    content_filter = ContentFilter(strategies)
+    filter_manifests(args.input_dir, args.output_dir, content_filter)
+
+
+if __name__ == "__main__":
+    main()

pipeline/segmentation/segment.sh

@@ -0,0 +1,13 @@
+#! /usr/bin/bash
+
+manifest_dir=${1%/}
+output_dir=${2%/}
+
+for manifest in ${manifest_dir}/*.jsonl; do
+  id=$(basename "${manifest}" .jsonl)
+  echo "Process $id"
+
+  python segment_from_manifests.py \
+    -m $manifest \
+    -o $output_dir
+done

pipeline/segmentation/segment_from_list.sh

@@ -0,0 +1,12 @@
+#! /usr/bin/bash
+
+output_dir=${2%/}
+
+while IFS= read -r manifest; do
+  id=$(basename "${manifest}" .jsonl)
+  echo "Process $id"
+
+  python segment_from_manifests.py \
+    -m $manifest \
+    -o $output_dir
+done < $1

pipeline/segmentation/segment_from_manifests.py

@@ -0,0 +1,52 @@
+import argparse
+import json
+import os
+
+import sox
+
+
+def main(args):
+    os.makedirs(args.output_dir, exist_ok=True)
+    audio_id = (
+        os.path.basename(args.manifest_filepath)
+        .replace("_manifest", "")
+        .replace("filtered_", "")
+        .split(".")[0]
+    )
+    segment_dir = os.path.join(args.output_dir, audio_id)
+    os.makedirs(segment_dir, exist_ok=True)
+    text_file_path = os.path.join(segment_dir, f"{audio_id}.trans.txt")
+
+    with open(args.manifest_filepath, "r") as reader, open(
+        text_file_path, "w"
+    ) as writer:
+        for i, line in enumerate(reader):
+            segment_id = f"{audio_id}-{i}"
+            line = json.loads(line)
+
+            audio_filepath = line["audio_filepath"]
+            audio_start_sec = line["audio_start_sec"]
+            audio_end_sec = audio_start_sec + line["duration"]
+
+            output_file = os.path.join(segment_dir, f"{segment_id}.wav")
+            tfm = sox.Transformer()
+            tfm.trim(audio_start_sec, audio_end_sec)
+            tfm.build_file(audio_filepath, output_file)
+
+            text = line["text"]
+            writer.write(segment_id + " " + text + "\n")
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description="Segment long audio files")
+    parser.add_argument(
+        "-m", "--manifest-filepath", type=str, help="Path to manifest file "
+    )
+    parser.add_argument(
+        "-o",
+        "--output-dir",
+        type=str,
+        help="Output directory to store segmented audio files",
+    )
+    args = parser.parse_args()
+    main(args)

pipeline/utils/force_alignment/README.md

@@ -0,0 +1,49 @@
+# Force Alignment for GigaSpeech 2
+Mainly from [fairseq](https://github.com/facebookresearch/fairseq/tree/main/examples/mms/data_prep), modified for force alignment in GigaSpeech 2
+# Data Preparation 
+
+We describe the process of aligning long audio files with their transcripts and generating shorter audio segments below. 
+
+- Step 1:  Download and install torchaudio>=2.1.0 . We have open sourced the CTC forced alignment algorithm described in our paper via [torchaudio](https://github.com/pytorch/audio/pull/3348). 
+  ```
+  pip install --pre torchaudio --index-url https://download.pytorch.org/whl/nightly/cu118
+  ```
+  
+- Step 2: Download [uroman](https://github.com/isi-nlp/uroman) from Github. It is a universal romanizer which converts text in any script to the Latin alphabet. Use [this link](https://www.isi.edu/~ulf/uroman.html) to try their web interface.  
+  ```
+  git clone git@github.com:isi-nlp/uroman.git
+  ```
+  
+- Step 3: Install a few other dependencies 
+  ```
+  pip install sox 
+  pip install dataclasses 
+  ```  
+
+- Step 4: Create a text file containing the transcript for a (long) audio file. Each line in the text file will correspond to a separate audio segment that will be generated upon alignment.
+
+  Example content of the input text file :
+  ```
+  Text of the desired first segment
+  Text of the desired second segment
+  Text of the desired third segment
+  ```
+
+- Step 5: Run forced alignment and segment the audio file into shorter segments. 
+  ```
+  python align_and_segment.py --audio /path/to/audio.wav --textfile /path/to/textfile --lang <iso> --outdir /path/to/output --uroman /path/to/uroman/bin 
+  ```
+
+  The above code  will generated the audio segments under output directory based on the content of each line in the input text file. The `manifest.json` file consisting of the of segmented audio filepaths and their corresponding transcripts. 
+
+  ```
+  > head /path/to/output/manifest.json 
+
+  {"audio_start_sec": 0.0, "audio_filepath": "/path/to/output/segment1.flac", "duration": 6.8, "text": "she wondered afterwards how she could have spoken with that hard serenity how she could have", "normalized_text": "she wondered afterwards how she could have spoken with that hard serenity how she could have", "uroman_tokens": "s h e w o n d e r e d a f t e r w a r d s h o w s h e c o u l d h a v e s p o k e n w i t h t h a t h a r d s e r e n i t y h o w s h e c o u l d h a v e"}
+  {"audio_start_sec": 6.8, "audio_filepath": "/path/to/output/segment2.flac", "duration": 5.3, "text": "gone steadily on with story after story poem after poem till", "normalized_text": "gone steadily on with story after story poem after poem till", "uroman_tokens": "g o n e s t e a d i l y o n w i t h s t o r y a f t e r s t o r y p o e m a f t e r p o e m t i l l"}
+  {"audio_start_sec": 12.1, "audio_filepath": "/path/to/output/segment3.flac", "duration": 5.9, "text": "allan's grip on her hands relaxed and he fell into a heavy tired sleep", "normalized_text": "allan's grip on her hands relaxed and he fell into a heavy tired sleep", "uroman_tokens": "a l l a n ' s g r i p o n h e r h a n d s r e l a x e d a n d h e f e l l i n t o a h e a v y t i r e d s l e e p"}
+  ```
+
+  To visualize the segmented audio files, [Speech Data Explorer](https://github.com/NVIDIA/NeMo/tree/main/tools/speech_data_explorer) tool from NeMo toolkit can be used.  
+
+  As our alignment model outputs uroman tokens for input audio in any language, it also works with non-english audio and their corresponding transcripts. 

pipeline/utils/force_alignment/align.py

@@ -0,0 +1,189 @@
+import argparse
+import json
+import os
+
+import sox
+import torch
+import torchaudio
+import torchaudio.functional as F
+
+from align_utils import (get_spans, get_uroman_tokens, load_model_dict,
+                         merge_repeats, time_to_frame)
+from text_normalization import text_normalize
+
+SAMPLING_FREQ = 16000
+EMISSION_INTERVAL = 30
+DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+
+
+def generate_emissions(model, audio_file):
+    waveform, _ = torchaudio.load(audio_file)  # waveform: channels X T
+    waveform = waveform.to(DEVICE)
+    total_duration = sox.file_info.duration(audio_file)
+
+    audio_sf = sox.file_info.sample_rate(audio_file)
+    assert audio_sf == SAMPLING_FREQ
+
+    emissions_arr = []
+    with torch.inference_mode():
+        i = 0
+        while i < total_duration:
+            segment_start_time, segment_end_time = (i, i + EMISSION_INTERVAL)
+
+            context = EMISSION_INTERVAL * 0.1
+            input_start_time = max(segment_start_time - context, 0)
+            input_end_time = min(segment_end_time + context, total_duration)
+            waveform_split = waveform[
+                :,
+                int(SAMPLING_FREQ * input_start_time) : int(
+                    SAMPLING_FREQ * (input_end_time)
+                ),
+            ]
+
+            model_outs, _ = model(waveform_split)
+            emissions_ = model_outs[0]
+            emission_start_frame = time_to_frame(segment_start_time)
+            emission_end_frame = time_to_frame(segment_end_time)
+            offset = time_to_frame(input_start_time)
+
+            emissions_ = emissions_[
+                emission_start_frame - offset : emission_end_frame - offset, :
+            ]
+            emissions_arr.append(emissions_)
+            i += EMISSION_INTERVAL
+
+    emissions = torch.cat(emissions_arr, dim=0).squeeze()
+    emissions = torch.log_softmax(emissions, dim=-1)
+
+    stride = float(waveform.size(1) * 1000 / emissions.size(0) / SAMPLING_FREQ)
+
+    return emissions, stride
+
+
+def get_alignments(
+    audio_file,
+    tokens,
+    model,
+    dictionary,
+    use_star,
+):
+    # Generate emissions
+    emissions, stride = generate_emissions(model, audio_file)
+    T, N = emissions.size()
+    if use_star:
+        emissions = torch.cat([emissions, torch.zeros(T, 1).to(DEVICE)], dim=1)
+
+    # Force Alignment
+    if tokens:
+        token_indices = [
+            dictionary[c] for c in " ".join(tokens).split(" ") if c in dictionary
+        ]
+    else:
+        print(f"Empty transcript!!!!! for audio file {audio_file}")
+        token_indices = []
+
+    blank = dictionary["<blank>"]
+
+    targets = torch.tensor(token_indices, dtype=torch.int32).to(DEVICE)
+
+    input_lengths = torch.tensor(emissions.shape[0]).unsqueeze(-1)
+    target_lengths = torch.tensor(targets.shape[0]).unsqueeze(-1)
+    path, _ = F.forced_align(
+        emissions.unsqueeze(0),
+        targets.unsqueeze(0),
+        input_lengths,
+        target_lengths,
+        blank=blank,
+    )
+    path = path.squeeze().to("cpu").tolist()
+
+    segments = merge_repeats(path, {v: k for k, v in dictionary.items()})
+    return segments, stride
+
+
+def main(args):
+    raw_transcripts = []
+    with open(args.text_filepath) as f:
+        raw_transcripts = [line.strip() for line in f]
+    print("Read {} lines from {}".format(len(raw_transcripts), args.text_filepath))
+
+    transcripts = []
+    norm_transcripts = []
+    for line in raw_transcripts:
+        transcript, norm_transcript = text_normalize(line.strip(), args.lang)
+        if len(norm_transcript) > 0:
+            transcripts.append(transcript)
+            norm_transcripts.append(norm_transcript)
+    tokens = get_uroman_tokens(norm_transcripts, args.uroman_path, args.lang)
+
+    model, dictionary = load_model_dict()
+    model = model.to(DEVICE)
+    if args.use_star:
+        dictionary["<star>"] = len(dictionary)
+        tokens = ["<star>"] + tokens
+        transcripts = ["<star>"] + transcripts
+        norm_transcripts = ["<star>"] + norm_transcripts
+
+    segments, stride = get_alignments(
+        args.audio_filepath,
+        tokens,
+        model,
+        dictionary,
+        args.use_star,
+    )
+    # Get spans of each line in input text file
+    spans = get_spans(tokens, segments, stride)
+
+    audio_id = os.path.basename(args.audio_filepath).split(".")[0]
+    os.makedirs(args.output_dir, exist_ok=True)
+    with open(f"{args.output_dir}/{audio_id}_manifest.jsonl", "w") as f:
+        for i, t in enumerate(norm_transcripts):
+            span = spans[i]
+            seg_start_idx = span[0].start
+            seg_end_idx = span[-1].end
+
+            audio_start_sec = seg_start_idx * stride / 1000
+            audio_end_sec = seg_end_idx * stride / 1000
+
+            sample = {
+                "audio_filepath": args.audio_filepath,
+                "audio_start_sec": audio_start_sec,
+                "duration": audio_end_sec - audio_start_sec,
+                "text": transcripts[i],
+            }
+            f.write(json.dumps(sample) + "\n")
+
+    return segments, stride
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description="Align and segment long audio files")
+    parser.add_argument(
+        "-a", "--audio-filepath", type=str, help="Path to input audio file"
+    )
+    parser.add_argument(
+        "-t", "--text-filepath", type=str, help="Path to input text file "
+    )
+    parser.add_argument(
+        "-l", "--lang", type=str, default="eng", help="ISO code of the language"
+    )
+    parser.add_argument(
+        "-u", "--uroman-path", type=str, default="eng", help="Location to uroman/bin"
+    )
+    parser.add_argument(
+        "-s",
+        "--use-star",
+        action="store_true",
+        help="Use star at the start of transcript",
+    )
+    parser.add_argument(
+        "-o",
+        "--output-dir",
+        type=str,
+        help="Output directory to store segmented audio files",
+    )
+    print("Using torch version:", torch.__version__)
+    print("Using torchaudio version:", torchaudio.__version__)
+    print("Using device: ", DEVICE)
+    args = parser.parse_args()
+    main(args)

pipeline/utils/force_alignment/align_utils.py

@@ -0,0 +1,194 @@
+import math
+import os
+import re
+import tempfile
+from dataclasses import dataclass
+
+import torch
+from torchaudio.models import wav2vec2_model
+
+# iso codes with specialized rules in uroman
+special_isos_uroman = "ara, bel, bul, deu, ell, eng, fas, grc, ell, eng, heb, kaz, kir, lav, lit, mkd, mkd2, oss, pnt, pus, rus, srp, srp2, tur, uig, ukr, yid".split(
+    ","
+)
+special_isos_uroman = [i.strip() for i in special_isos_uroman]
+
+
+def normalize_uroman(text):
+    text = text.lower()
+    text = re.sub("([^a-z' ])", " ", text)
+    text = re.sub(" +", " ", text)
+    return text.strip()
+
+
+def get_uroman_tokens(norm_transcripts, uroman_root_dir, iso=None):
+    tf = tempfile.NamedTemporaryFile()
+    tf2 = tempfile.NamedTemporaryFile()
+    with open(tf.name, "w") as f:
+        for t in norm_transcripts:
+            f.write(t + "\n")
+
+    assert os.path.exists(f"{uroman_root_dir}/uroman.pl"), "uroman not found"
+    cmd = f"perl {uroman_root_dir}/uroman.pl"
+    if iso in special_isos_uroman:
+        cmd += f" -l {iso} "
+    cmd += f" < {tf.name} > {tf2.name}"
+    os.system(cmd)
+    outtexts = []
+    with open(tf2.name) as f:
+        for line in f:
+            line = " ".join(line.strip())
+            line = re.sub(r"\s+", " ", line).strip()
+            outtexts.append(line)
+    assert len(outtexts) == len(norm_transcripts)
+    uromans = []
+    for ot in outtexts:
+        uromans.append(normalize_uroman(ot))
+    return uromans
+
+
+@dataclass
+class Segment:
+    label: str
+    start: int
+    end: int
+
+    def __repr__(self):
+        return f"{self.label}: [{self.start:5d}, {self.end:5d})"
+
+    @property
+    def length(self):
+        return self.end - self.start
+
+
+def merge_repeats(path, idx_to_token_map):
+    i1, i2 = 0, 0
+    segments = []
+    while i1 < len(path):
+        while i2 < len(path) and path[i1] == path[i2]:
+            i2 += 1
+        segments.append(Segment(idx_to_token_map[path[i1]], i1, i2 - 1))
+        i1 = i2
+    return segments
+
+
+def time_to_frame(time):
+    stride_msec = 20
+    frames_per_sec = 1000 / stride_msec
+    return int(time * frames_per_sec)
+
+
+def load_model_dict():
+    model_path_name = "/tmp/ctc_alignment_mling_uroman_model.pt"
+
+    print("Downloading model and dictionary...")
+    if os.path.exists(model_path_name):
+        print("Model path already exists. Skipping downloading....")
+    else:
+        torch.hub.download_url_to_file(
+            "https://dl.fbaipublicfiles.com/mms/torchaudio/ctc_alignment_mling_uroman/model.pt",
+            model_path_name,
+        )
+        assert os.path.exists(model_path_name)
+    state_dict = torch.load(model_path_name, map_location="cpu")
+
+    model = wav2vec2_model(
+        extractor_mode="layer_norm",
+        extractor_conv_layer_config=[
+            (512, 10, 5),
+            (512, 3, 2),
+            (512, 3, 2),
+            (512, 3, 2),
+            (512, 3, 2),
+            (512, 2, 2),
+            (512, 2, 2),
+        ],
+        extractor_conv_bias=True,
+        encoder_embed_dim=1024,
+        encoder_projection_dropout=0.0,
+        encoder_pos_conv_kernel=128,
+        encoder_pos_conv_groups=16,
+        encoder_num_layers=24,
+        encoder_num_heads=16,
+        encoder_attention_dropout=0.0,
+        encoder_ff_interm_features=4096,
+        encoder_ff_interm_dropout=0.1,
+        encoder_dropout=0.0,
+        encoder_layer_norm_first=True,
+        encoder_layer_drop=0.1,
+        aux_num_out=31,
+    )
+    model.load_state_dict(state_dict)
+    model.eval()
+
+    dict_path_name = "/tmp/ctc_alignment_mling_uroman_model.dict"
+    if os.path.exists(dict_path_name):
+        print("Dictionary path already exists. Skipping downloading....")
+    else:
+        torch.hub.download_url_to_file(
+            "https://dl.fbaipublicfiles.com/mms/torchaudio/ctc_alignment_mling_uroman/dictionary.txt",
+            dict_path_name,
+        )
+        assert os.path.exists(dict_path_name)
+    dictionary = {}
+    with open(dict_path_name) as f:
+        dictionary = {l.strip(): i for i, l in enumerate(f.readlines())}
+
+    return model, dictionary
+
+
+def get_spans(tokens, segments, stride):
+    ltr_idx = 0
+    tokens_idx = 0
+    intervals = []
+    start, end = (0, 0)
+    sil = "<blank>"
+    for seg_idx, seg in enumerate(segments):
+        if tokens_idx == len(tokens):
+            assert seg_idx == len(segments) - 1
+            assert seg.label == "<blank>"
+            continue
+        cur_token = tokens[tokens_idx].split(" ")
+        ltr = cur_token[ltr_idx]
+        if seg.label == "<blank>":
+            continue
+        assert seg.label == ltr
+        if (ltr_idx) == 0:
+            start = seg_idx
+        if ltr_idx == len(cur_token) - 1:
+            ltr_idx = 0
+            tokens_idx += 1
+            intervals.append((start, seg_idx))
+            while tokens_idx < len(tokens) and len(tokens[tokens_idx]) == 0:
+                intervals.append((seg_idx, seg_idx))
+                tokens_idx += 1
+        else:
+            ltr_idx += 1
+    spans = []
+    for idx, (start, end) in enumerate(intervals):
+        span = segments[start : end + 1]
+        # don't need the segments to be connected
+        if start > 0:
+            prev_seg = segments[start - 1]
+            if prev_seg.label == sil:
+                pad_start = (
+                    prev_seg.start
+                    if (idx == 0)
+                    else int((prev_seg.start + prev_seg.end) / 2)
+                )
+                pad_start = max(
+                    pad_start, math.floor(prev_seg.end - 0.1 * 1000 / stride)
+                )
+                span = [Segment(sil, pad_start, span[0].start)] + span
+        if end + 1 < len(segments):
+            next_seg = segments[end + 1]
+            if next_seg.label == sil:
+                pad_end = (
+                    next_seg.end
+                    if (idx == len(intervals) - 1)
+                    else math.floor((next_seg.start + next_seg.end) / 2)
+                )
+                pad_end = min(pad_end, math.floor(next_seg.start + 0.5 * 1000 / stride))
+                span = span + [Segment(sil, span[-1].end, pad_end)]
+        spans.append(span)
+    return spans

pipeline/utils/force_alignment/norm_config.py

@@ -0,0 +1,277 @@
+import os
+import re
+
+colon = ":"
+comma = ","
+exclamation_mark = "!"
+period = re.escape(".")
+question_mark = re.escape("?")
+semicolon = ";"
+
+dash = re.escape("-")
+underscore = "_"
+
+left_curly_bracket = "{"
+right_curly_bracket = "}"
+quotation_mark = '"'
+single_quotation_mark = "'"
+
+basic_punc = (
+    period
+    + question_mark
+    + comma
+    + colon
+    + exclamation_mark
+    + left_curly_bracket
+    + right_curly_bracket
+    + dash
+    + underscore
+    + single_quotation_mark
+)
+
+# General punc unicode block (0x2000-0x206F)
+zero_width_space = r"\u200B"
+zero_width_nonjoiner = r"\u200C"
+left_to_right_mark = r"\u200E"
+right_to_left_mark = r"\u200F"
+left_to_right_embedding = r"\u202A"
+pop_directional_formatting = r"\u202C"
+
+# Here are some commonly ill-typed versions of apostrophe
+right_single_quotation_mark = r"\u2019"
+left_single_quotation_mark = r"\u2018"
+
+# Language specific definitions
+# Spanish
+inverted_exclamation_mark = r"\u00A1"
+inverted_question_mark = r"\u00BF"
+
+
+# Hindi
+hindi_danda = "\u0964"
+
+# Egyptian Arabic
+# arabic_percent = r"\u066A"
+arabic_comma = r"\u060C"
+arabic_question_mark = r"\u061F"
+arabic_semicolon = r"\u061B"
+arabic_diacritics = r"\u064B-\u0652"
+
+
+arabic_subscript_alef_and_inverted_damma = r"\u0656-\u0657"
+
+
+# Chinese
+full_stop = r"\u3002"
+full_comma = r"\uFF0C"
+full_exclamation_mark = r"\uFF01"
+full_question_mark = r"\uFF1F"
+full_semicolon = r"\uFF1B"
+full_colon = r"\uFF1A"
+full_parentheses = r"\uFF08\uFF09"
+quotation_mark_horizontal = r"\u300C-\u300F"
+quotation_mark_vertical = r"\uFF41-\uFF44"
+title_marks = r"\u3008-\u300B"
+wavy_low_line = r"\uFE4F"
+ellipsis = r"\u22EF"
+enumeration_comma = r"\u3001"
+hyphenation_point = r"\u2027"
+forward_slash = r"\uFF0F"
+wavy_dash = r"\uFF5E"
+box_drawings_light_horizontal = r"\u2500"
+fullwidth_low_line = r"\uFF3F"
+chinese_punc = (
+    full_stop
+    + full_comma
+    + full_exclamation_mark
+    + full_question_mark
+    + full_semicolon
+    + full_colon
+    + full_parentheses
+    + quotation_mark_horizontal
+    + quotation_mark_vertical
+    + title_marks
+    + wavy_low_line
+    + ellipsis
+    + enumeration_comma
+    + hyphenation_point
+    + forward_slash
+    + wavy_dash
+    + box_drawings_light_horizontal
+    + fullwidth_low_line
+)
+
+# Armenian
+armenian_apostrophe = r"\u055A"
+emphasis_mark = r"\u055B"
+exclamation_mark = r"\u055C"
+armenian_comma = r"\u055D"
+armenian_question_mark = r"\u055E"
+abbreviation_mark = r"\u055F"
+armenian_full_stop = r"\u0589"
+armenian_punc = (
+    armenian_apostrophe
+    + emphasis_mark
+    + exclamation_mark
+    + armenian_comma
+    + armenian_question_mark
+    + abbreviation_mark
+    + armenian_full_stop
+)
+
+lesser_than_symbol = r"<"
+greater_than_symbol = r">"
+
+lesser_than_sign = r"\u003c"
+greater_than_sign = r"\u003e"
+
+nbsp_written_form = r"&nbsp"
+
+# Quotation marks
+left_double_quotes = r"\u201c"
+right_double_quotes = r"\u201d"
+left_double_angle = r"\u00ab"
+right_double_angle = r"\u00bb"
+left_single_angle = r"\u2039"
+right_single_angle = r"\u203a"
+low_double_quotes = r"\u201e"
+low_single_quotes = r"\u201a"
+high_double_quotes = r"\u201f"
+high_single_quotes = r"\u201b"
+
+all_punct_quotes = (
+    left_double_quotes
+    + right_double_quotes
+    + left_double_angle
+    + right_double_angle
+    + left_single_angle
+    + right_single_angle
+    + low_double_quotes
+    + low_single_quotes
+    + high_double_quotes
+    + high_single_quotes
+    + right_single_quotation_mark
+    + left_single_quotation_mark
+)
+mapping_quotes = (
+    "["
+    + high_single_quotes
+    + right_single_quotation_mark
+    + left_single_quotation_mark
+    + "]"
+)
+
+
+# Digits
+english_digits = r"\u0030-\u0039"
+bengali_digits = r"\u09e6-\u09ef"
+khmer_digits = r"\u17e0-\u17e9"
+devanagari_digits = r"\u0966-\u096f"
+oriya_digits = r"\u0b66-\u0b6f"
+extended_arabic_indic_digits = r"\u06f0-\u06f9"
+kayah_li_digits = r"\ua900-\ua909"
+fullwidth_digits = r"\uff10-\uff19"
+malayam_digits = r"\u0d66-\u0d6f"
+myanmar_digits = r"\u1040-\u1049"
+roman_numeral = r"\u2170-\u2179"
+nominal_digit_shapes = r"\u206f"
+
+# Load punctuations from MMS-lab data
+with open(f"{os.path.dirname(__file__)}/punctuations.lst", "r") as punc_f:
+    punc_list = punc_f.readlines()
+
+punct_pattern = r""
+for punc in punc_list:
+    # the first character in the tab separated line is the punc to be removed
+    punct_pattern += re.escape(punc.split("\t")[0])
+
+shared_digits = (
+    english_digits
+    + bengali_digits
+    + khmer_digits
+    + devanagari_digits
+    + oriya_digits
+    + extended_arabic_indic_digits
+    + kayah_li_digits
+    + fullwidth_digits
+    + malayam_digits
+    + myanmar_digits
+    + roman_numeral
+    + nominal_digit_shapes
+)
+
+shared_punc_list = (
+    basic_punc
+    + all_punct_quotes
+    + greater_than_sign
+    + lesser_than_sign
+    + inverted_question_mark
+    + full_stop
+    + semicolon
+    + armenian_punc
+    + inverted_exclamation_mark
+    + arabic_comma
+    + enumeration_comma
+    + hindi_danda
+    + quotation_mark
+    + arabic_semicolon
+    + arabic_question_mark
+    + chinese_punc
+    + punct_pattern
+)
+
+shared_mappping = {
+    lesser_than_symbol: "",
+    greater_than_symbol: "",
+    nbsp_written_form: "",
+    r"(\S+)" + mapping_quotes + r"(\S+)": r"\1'\2",
+}
+
+shared_deletion_list = (
+    left_to_right_mark
+    + zero_width_nonjoiner
+    + arabic_subscript_alef_and_inverted_damma
+    + zero_width_space
+    + arabic_diacritics
+    + pop_directional_formatting
+    + right_to_left_mark
+    + left_to_right_embedding
+)
+
+norm_config = {
+    "*": {
+        "lower_case": True,
+        "punc_set": shared_punc_list,
+        "del_set": shared_deletion_list,
+        "mapping": shared_mappping,
+        "digit_set": shared_digits,
+        "unicode_norm": "NFKC",
+        "rm_diacritics": False,
+    }
+}
+
+# =============== Mongolian ===============#
+norm_config["mon"] = norm_config["*"].copy()
+# add soft hyphen to punc list to match with fleurs
+norm_config["mon"]["del_set"] += r"\u00AD"
+
+norm_config["khk"] = norm_config["mon"].copy()
+
+# =============== Hebrew ===============#
+norm_config["heb"] = norm_config["*"].copy()
+# add "HEBREW POINT" symbols to match with fleurs
+norm_config["heb"]["del_set"] += r"\u05B0-\u05BF\u05C0-\u05CF"
+
+# =============== Thai ===============#
+norm_config["tha"] = norm_config["*"].copy()
+# add "Zero width joiner" symbols to match with fleurs
+norm_config["tha"]["punc_set"] += r"\u200D"
+
+# =============== Arabic ===============#
+norm_config["ara"] = norm_config["*"].copy()
+norm_config["ara"]["mapping"]["ٱ"] = "ا"
+norm_config["arb"] = norm_config["ara"].copy()
+
+# =============== Javanese ===============#
+norm_config["jav"] = norm_config["*"].copy()
+norm_config["jav"]["rm_diacritics"] = True

pipeline/utils/force_alignment/punctuations.lst

@@ -0,0 +1,188 @@
+	7355	INVALID UNICODE	0x81
+	5265	INVALID UNICODE	0x90
+	75	INVALID UNICODE	0x8
+	31	INVALID UNICODE	0x8d
+”	3	INVALID UNICODE	0x94
+	2	INVALID UNICODE	0x8f
+	2	INVALID UNICODE	0x1a
+	1	INVALID UNICODE	0x9d
+“	1	INVALID UNICODE	0x93
+’	1	INVALID UNICODE	0x92
+	8647	INVALID UNICODE	0xe295
+	6650	INVALID UNICODE	0xf21d
+	6234	INVALID UNICODE	0xf62d
+	4815	INVALID UNICODE	0xf173
+	4789	INVALID UNICODE	0xe514
+	4409	INVALID UNICODE	0xe293
+	3881	INVALID UNICODE	0xf523
+	3788	INVALID UNICODE	0xe233
+	2448	INVALID UNICODE	0xf50f
+	2177	INVALID UNICODE	0xe232
+	1955	INVALID UNICODE	0xea7b
+	1926	INVALID UNICODE	0xf172
+	973	INVALID UNICODE	0xe290
+	972	INVALID UNICODE	0xf519
+	661	INVALID UNICODE	0xe292
+	591	INVALID UNICODE	0xe328
+	509	INVALID UNICODE	0xe2fa
+	458	INVALID UNICODE	0xe234
+	446	INVALID UNICODE	0xe043
+	419	INVALID UNICODE	0xe040
+	399	INVALID UNICODE	0xe2fb
+	387	INVALID UNICODE	0xe32b
+	381	INVALID UNICODE	0xe236
+	374	INVALID UNICODE	0xf511
+	314	INVALID UNICODE	0xe517
+	296	INVALID UNICODE	0xe2fe
+	293	INVALID UNICODE	0xe492
+	291	INVALID UNICODE	0xf52d
+	289	INVALID UNICODE	0xe2fc
+	195	INVALID UNICODE	0xf521
+	190	INVALID UNICODE	0xe516
+	182	INVALID UNICODE	0xe041
+	178	INVALID UNICODE	0xf529
+	113	INVALID UNICODE	0xe2f9
+	87	INVALID UNICODE	0xe2d9
+	78	INVALID UNICODE	0xe32a
+	76	INVALID UNICODE	0xe291
+	74	INVALID UNICODE	0xe296
+	66	INVALID UNICODE	0xe518
+	52	INVALID UNICODE	0xe32c
+	46	INVALID UNICODE	0xe2db
+	41	INVALID UNICODE	0xe231
+	34	INVALID UNICODE	0xf522
+	33	INVALID UNICODE	0xf518
+	32	INVALID UNICODE	0xf513
+	27	INVALID UNICODE	0xe32d
+	25	INVALID UNICODE	0xe32e
+	23	INVALID UNICODE	0xe06b
+	15	INVALID UNICODE	0xea01
+	12	INVALID UNICODE	0xe294
+	11	INVALID UNICODE	0xe203
+	8	INVALID UNICODE	0xf218
+	7	INVALID UNICODE	0xe070
+	7	INVALID UNICODE	0xe013
+	5	INVALID UNICODE	0xe2de
+	4	INVALID UNICODE	0xe493
+	3	INVALID UNICODE	0xf7e8
+	3	INVALID UNICODE	0xf7d0
+	3	INVALID UNICODE	0xe313
+	2	INVALID UNICODE	0xe329
+	2	INVALID UNICODE	0xe06d
+	2	INVALID UNICODE	0xe003
+	1	INVALID UNICODE	0xf50e
+	1	INVALID UNICODE	0xf171
+	1	INVALID UNICODE	0xe01d
+	71	NOMINAL DIGIT SHAPES	0x206f
+⁠	3	WORD JOINER	0x2060
+―	126545	HORIZONTAL BAR	0x2015
+־	1028	HEBREW PUNCTUATION MAQAF	0x5be
+)	98429	RIGHT PARENTHESIS	0x29
+]	27108	RIGHT SQUARE BRACKET	0x5d
+⌋	1567	RIGHT FLOOR	0x230b
+〕	97	RIGHT TORTOISE SHELL BRACKET	0x3015
+】	36	RIGHT BLACK LENTICULAR BRACKET	0x3011
+﴾	14	ORNATE LEFT PARENTHESIS	0xfd3e
+&	170517	AMPERSAND	0x26
+།	106330	TIBETAN MARK SHAD	0xf0d
+።	90203	ETHIOPIC FULL STOP	0x1362
+፥	60484	ETHIOPIC COLON	0x1365
+༌	60464	TIBETAN MARK DELIMITER TSHEG BSTAR	0xf0c
+။	51567	MYANMAR SIGN SECTION	0x104b
+/	46929	SOLIDUS	0x2f
+၊	38042	MYANMAR SIGN LITTLE SECTION	0x104a
+·	37985	MIDDLE DOT	0xb7
+‸	36310	CARET	0x2038
+*	34793	ASTERISK	0x2a
+۔	32432	ARABIC FULL STOP	0x6d4
+፤	31906	ETHIOPIC SEMICOLON	0x1364
+၏	21519	MYANMAR SYMBOL GENITIVE	0x104f
+។	20834	KHMER SIGN KHAN	0x17d4
+꓾	15773	LISU PUNCTUATION COMMA	0xa4fe
+᙮	13473	CANADIAN SYLLABICS FULL STOP	0x166e
+꤯	12892	KAYAH LI SIGN SHYA	0xa92f
+⵰	11478	TIFINAGH SEPARATOR MARK	0x2d70
+꓿	11118	LISU PUNCTUATION FULL STOP	0xa4ff
+॥	10763	DEVANAGARI DOUBLE DANDA	0x965
+؞	10403	ARABIC TRIPLE DOT PUNCTUATION MARK	0x61e
+၍	8936	MYANMAR SYMBOL COMPLETED	0x104d
+·	8431	GREEK ANO TELEIA	0x387
+†	7477	DAGGER	0x2020
+၌	6632	MYANMAR SYMBOL LOCATIVE	0x104c
+፣	5719	ETHIOPIC COMMA	0x1363
+៖	5528	KHMER SIGN CAMNUC PII KUUH	0x17d6
+꤮	4791	KAYAH LI SIGN CWI	0xa92e
+※	3439	REFERENCE MARK	0x203b
+፦	2727	ETHIOPIC PREFACE COLON	0x1366
+•	1749	BULLET	0x2022
+¶	1507	PILCROW SIGN	0xb6
+၎	1386	MYANMAR SYMBOL AFOREMENTIONED	0x104e
+﹖	1224	SMALL QUESTION MARK	0xfe56
+;	975	GREEK QUESTION MARK	0x37e
+…	827	HORIZONTAL ELLIPSIS	0x2026
+%	617	PERCENT SIGN	0x25
+・	468	KATAKANA MIDDLE DOT	0x30fb
+༎	306	TIBETAN MARK NYIS SHAD	0xf0e
+‡	140	DOUBLE DAGGER	0x2021
+#	137	NUMBER SIGN	0x23
+@	125	COMMERCIAL AT	0x40
+፡	121	ETHIOPIC WORDSPACE	0x1361
+៚	55	KHMER SIGN KOOMUUT	0x17da
+៕	49	KHMER SIGN BARIYOOSAN	0x17d5
+﹐	10	SMALL COMMA	0xfe50
+༅	6	TIBETAN MARK CLOSING YIG MGO SGAB MA	0xf05
+༄	6	TIBETAN MARK INITIAL YIG MGO MDUN MA	0xf04
+．	2	FULLWIDTH FULL STOP	0xff0e
+﹗	2	SMALL EXCLAMATION MARK	0xfe57
+﹕	2	SMALL COLON	0xfe55
+‰	2	PER MILLE SIGN	0x2030
+･	1	HALFWIDTH KATAKANA MIDDLE DOT	0xff65
+(	98504	LEFT PARENTHESIS	0x28
+[	27245	LEFT SQUARE BRACKET	0x5b
+⌊	1567	LEFT FLOOR	0x230a
+〔	95	LEFT TORTOISE SHELL BRACKET	0x3014
+【	36	LEFT BLACK LENTICULAR BRACKET	0x3010
+﴿	14	ORNATE RIGHT PARENTHESIS	0xfd3f
+_	4851	LOW LINE	0x5f
+$	72	DOLLAR SIGN	0x24
+€	14	EURO SIGN	0x20ac
+£	2	POUND SIGN	0xa3
+~	27462	TILDE	0x7e
+=	11450	EQUALS SIGN	0x3d
+|	8430	VERTICAL LINE	0x7c
+−	3971	MINUS SIGN	0x2212
+≫	1904	MUCH GREATER-THAN	0x226b
+≪	1903	MUCH LESS-THAN	0x226a
++	1450	PLUS SIGN	0x2b
+＜	345	FULLWIDTH LESS-THAN SIGN	0xff1c
+＞	344	FULLWIDTH GREATER-THAN SIGN	0xff1e
+¬	5	NOT SIGN	0xac
+×	4	MULTIPLICATION SIGN	0xd7
+→	2	RIGHTWARDS ARROW	0x2192
+᙭	537	CANADIAN SYLLABICS CHI SIGN	0x166d
+°	499	DEGREE SIGN	0xb0
+႟	421	MYANMAR SYMBOL SHAN EXCLAMATION	0x109f
+�	192	REPLACEMENT CHARACTER	0xfffd
+⌟	54	BOTTOM RIGHT CORNER	0x231f
+⌞	54	BOTTOM LEFT CORNER	0x231e
+©	2	COPYRIGHT SIGN	0xa9
+ 	40	NARROW NO-BREAK SPACE	0x202f
+ 	1	SIX-PER-EM SPACE	0x2006
+˜	40261	SMALL TILDE	0x2dc
+^	6469	CIRCUMFLEX ACCENT	0x5e
+¯	20	MACRON	0xaf
+ˇ	191442	CARON	0x2c7
+ⁿ	38144	SUPERSCRIPT LATIN SMALL LETTER N	0x207f
+ـ	9440	ARABIC TATWEEL	0x640
+ๆ	6766	THAI CHARACTER MAIYAMOK	0xe46
+ៗ	3310	KHMER SIGN LEK TOO	0x17d7
+々	678	IDEOGRAPHIC ITERATION MARK	0x3005
+ໆ	430	LAO KO LA	0xec6
+ー	319	KATAKANA-HIRAGANA PROLONGED SOUND MARK	0x30fc
+ⁱ	137	SUPERSCRIPT LATIN SMALL LETTER I	0x2071
+৷	11056	BENGALI CURRENCY NUMERATOR FOUR	0x9f7
+⅓	26	VULGAR FRACTION ONE THIRD	0x2153
+½	26	VULGAR FRACTION ONE HALF	0xbd
+¼	4	VULGAR FRACTION ONE QUARTER	0xbc
+⅟	1	FRACTION NUMERATOR ONE	0x215f
+⁄	57	FRACTION SLASH	0x2044

pipeline/utils/force_alignment/text_normalization.py

@@ -0,0 +1,92 @@
+import json
+import re
+import unicodedata
+
+from norm_config import norm_config
+
+
+def text_normalize(
+    text, iso_code, lower_case=True, remove_numbers=True, remove_brackets=False
+):
+    """Given a text, normalize it by changing to lower case, removing punctuations, removing words that only contain digits and removing extra spaces
+
+    Args:
+        text: The string to be normalized
+        remove_numbers: Boolean flag to specify if words containing only digits should be removed
+
+    Returns:
+        normalized_text: the string after all normalization
+
+    """
+
+    config = norm_config.get(iso_code, norm_config["*"])
+
+    for field in [
+        "lower_case",
+        "punc_set",
+        "del_set",
+        "mapping",
+        "digit_set",
+        "unicode_norm",
+    ]:
+        if field not in config:
+            config[field] = norm_config["*"][field]
+
+    text = unicodedata.normalize(config["unicode_norm"], text)
+
+    # Convert to lower case
+    if config["lower_case"] and lower_case:
+        text = text.lower()
+
+    # brackets
+    # always text inside brackets with numbers in them. Usually corresponds to "(Sam 23:17)"
+    text = re.sub(r"\([^\)]*\d[^\)]*\)", " ", text)
+    if remove_brackets:
+        text = re.sub(r"\([^\)]*\)", " ", text)
+
+    # Apply mappings
+    for old, new in config["mapping"].items():
+        text = re.sub(old, new, text)
+
+    # Replace punctutations with space
+    punct_pattern = r"[" + config["punc_set"]
+    punct_pattern += "]"
+    text = re.sub(punct_pattern, " ", text)
+
+    # remove characters in delete list
+    delete_patten = r"[" + config["del_set"] + "]"
+    text = re.sub(delete_patten, "", text)
+
+    # Remove words containing only digits
+    # We check for 3 cases  a)text starts with a number b) a number is present somewhere in the middle of the text c) the text ends with a number
+    # For each case we use lookaround regex pattern to see if the digit pattern in preceded and followed by whitespaces, only then we replace the numbers with space
+    # The lookaround enables overlapping pattern matches to be replaced
+
+    if remove_numbers:
+        digits_pattern = "[" + config["digit_set"]
+        digits_pattern += "]+"
+        complete_digit_pattern = (
+            r"^"
+            + digits_pattern
+            + "(?=\s)|(?<=\s)"
+            + digits_pattern
+            + "(?=\s)|(?<=\s)"
+            + digits_pattern
+            + "$"
+        )
+        normalized_text = re.sub(complete_digit_pattern, " ", text)
+
+    if config["rm_diacritics"]:
+        from unidecode import unidecode
+
+        normalized_text = unidecode(normalized_text)
+
+    # Remove extra spaces
+    text = re.sub(r"\s+", " ", text).strip()
+    normalized_text = re.sub(r"\s+", " ", normalized_text).strip()
+
+    # Remove repeating symbols
+    repeat_pattern = r"(.)\1{4,}"
+    normalized_text = re.sub(repeat_pattern, r"\1", normalized_text)
+
+    return text, normalized_text

pipeline/utils/textgrid2jsonl.py

@@ -0,0 +1,22 @@
+import json
+import os
+import textgrid
+from lhotse.utils import add_durations 
+
+
+for i in os.listdir("textgrid"):
+    tg_path = os.path.join("textgrid", i)
+    tg = textgrid.TextGrid()
+    tg.read(tg_path)
+
+    manifest_path = i.replace(".TextGrid", "_manifest.jsonl")
+    with open("mms/" + manifest_path, "w") as f:
+        for interval in tg.tiers[0]:
+            if len(interval.mark) == 0:
+                continue
+            line = {}
+            line["audio_filepath"] = os.path.join("/data/shared/Thai_test_merge/wav", i.replace(".TextGrid", ".wav"))
+            line["audio_start_sec"] = interval.minTime
+            line["duration"] = add_durations(interval.maxTime, -interval.minTime, sampling_rate=16000)
+            line["text"] = interval.mark
+            f.write(json.dumps(line) + "\n")

pipeline/utils/uroman/.gitignore

@@ -0,0 +1,35 @@
+!Build/
+.last_cover_stats
+/META.yml
+/META.json
+/MYMETA.*
+*.o
+*.pm.tdy
+*.bs
+
+# Devel::Cover
+cover_db/
+
+# Devel::NYTProf
+nytprof.out
+
+# Dizt::Zilla
+/.build/
+
+# Module::Build
+_build/
+Build
+Build.bat
+
+# Module::Install
+inc/
+
+# ExtUtils::MakeMaker
+/blib/
+/_eumm/
+/*.gz
+/Makefile
+/Makefile.old
+/MANIFEST.bak
+/pm_to_blib
+/*.zip

pipeline/utils/uroman/LICENSE.txt

@@ -0,0 +1,11 @@
+Copyright (C) 2015-2020 Ulf Hermjakob, USC Information Sciences Institute
+
+Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
+
+Any publication of projects using uroman shall acknowledge its use: "This project uses the universal romanizer software 'uroman' written by Ulf Hermjakob, USC Information Sciences Institute (2015-2020)".
+Bibliography: Ulf Hermjakob, Jonathan May, and Kevin Knight. 2018. Out-of-the-box universal romanization tool uroman. In Proceedings of the 56th Annual Meeting of Association for Computational Linguistics, Demo Track.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+

pipeline/utils/uroman/README.md

@@ -0,0 +1,165 @@
+# uroman
+
+*uroman* is a *universal romanizer*. It converts text in any script to the Latin alphabet.
+
+Version: 1.2.8
+Release date: April 23, 2021  
+Author: Ulf Hermjakob, USC Information Sciences Institute  
+
+
+### Usage
+```bash
+$ uroman.pl [-l <lang-code>] [--chart] [--no-cache] < STDIN
+       where the optional <lang-code> is a 3-letter languages code, e.g. ara, bel, bul, deu, ell, eng, fas,
+            grc, ell, eng, heb, kaz, kir, lav, lit, mkd, mkd2, oss, pnt, pus, rus, srp, srp2, tur, uig, ukr, yid.
+       --chart specifies chart output (in JSON format) to represent alternative romanizations.
+       --no-cache disables caching.
+```
+### Examples
+```bash
+$ bin/uroman.pl < text/zho.txt
+$ bin/uroman.pl -l tur < text/tur.txt
+$ bin/uroman.pl -l heb --chart < text/heb.txt
+$ bin/uroman.pl < test/multi-script.txt > test/multi-script.uroman.txt
+```
+
+Identifying the input as Arabic, Belarusian, Bulgarian, English, Farsi, German,
+Ancient Greek, Modern Greek, Pontic Greek, Hebrew, Kazakh, Kyrgyz, Latvian,
+Lithuanian, North Macedonian, Russian, Serbian, Turkish, Ukrainian, Uyghur or 
+Yiddish will improve romanization for those languages as some letters in those 
+languages have different sound values from other languages using the same script 
+(French, Russian, Hebrew respectively).
+No effect for other languages in this version.
+
+### Bibliography
+Ulf Hermjakob, Jonathan May, and Kevin Knight. 2018. Out-of-the-box universal romanization tool uroman. In Proceedings of the 56th Annual Meeting of Association for Computational Linguistics, Demo Track. ACL-2018 Best Demo Paper Award. [Paper in ACL Anthology](https://www.aclweb.org/anthology/P18-4003) | [Poster](https://www.isi.edu/~ulf/papers/poster-uroman-acl2018.pdf) | [BibTex](https://www.aclweb.org/anthology/P18-4003.bib)
+
+### Change History
+Changes in version 1.2.8
+ * Updated to Unicode 13.0 (2021), which supports several new scripts (10% larger UnicodeData.txt).
+ * Improved support for Georgian.
+ * Preserve various symbols (as opposed to mapping to the symbols' names).
+ * Various small improvements.
+
+Changes in version 1.2.7
+ * Improved support for Pashto.
+
+Changes in version 1.2.6
+ * Improved support for Ukrainian, Russian and Ogham (ancient Irish script).
+ * Added support for English Braille.
+ * Added alternative Romanization for North Macedonian and Serbian (mkd2/srp2)
+   reflecting a casual style that many native speakers of those languages use
+   when writing text in Latin script, e.g. non-accented single letters (e.g. "s")
+   rather than phonetically motivated combinations of letters (e.g. "sh").
+ * When a line starts with "::lcode xyz ", the new uroman version will switch to
+   that language for that line. This is used for the new reference test file.
+ * Various small improvements.
+
+Changes in version 1.2.5
+ * Improved support for Armenian and eight languages using Cyrillic scripts.
+   -- For Serbian and Macedonian, which are often written in both Cyrillic
+      and Latin scripts, uroman will map both official versions to the same
+      romanized text, e.g. both "Ниш" and "Niš" will be mapped to "Nish" (which
+      properly reflects the pronunciation of the city's name).
+      For both Serbian and Macedonian, casual writers often use a simplified
+      Latin form without diacritics, e.g. "s" to represent not only Cyrillic "с"
+      and Latin "s", but also "ш" or "š", even if this conflates "s" and "sh" and
+      other such pairs. The casual romanization can be simulated by using
+      alternative uroman language codes "srp2" and "mkd2", which romanize
+      both "Ниш" and "Niš" to "Nis" to reflect the casual Latin spelling.
+ * Various small improvements.
+
+Changes in version 1.2.4
+  * Bug-fix that generated two emtpy lines for each empty line in cache mode.
+
+Changes in version 1.2
+ * Run-time improvement based on (1) token-based caching and (2) shortcut 
+   romanization (identity) of ASCII strings for default 1-best (non-chart) 
+   output. Speed-up by a factor of 10 for Bengali and Uyghur on medium and 
+   large size texts.
+ * Incremental improvements for Farsi, Amharic, Russian, Hebrew and related
+   languages.
+ * Richer lattice structure (more alternatives) for "Romanization" of English
+   to support better matching to romanizations of other languages.
+   Changes output only when --chart option is specified. No change in output for
+   default 1-best output, which for ASCII characters is always the input string.
+
+Changes in version 1.1 (major upgrade)
+ * Offers chart output (in JSON format) to represent alternative romanizations.
+   -- Location of first character is defined to be "line: 1, start:0, end:0".
+ * Incremental improvements of Hebrew and Greek romanization; Chinese numbers.
+ * Improved web-interface at http://www.isi.edu/~ulf/uroman.html
+   -- Shows corresponding original and romanization text in red
+      when hovering over a text segment.
+   -- Shows alternative romanizations when hovering over romanized text
+      marked by dotted underline.
+   -- Added right-to-left script detection and improved display for right-to-left
+      script text (as determined line by line).
+   -- On-page support for some scripts that are often not pre-installed on users'
+      computers (Burmese, Egyptian, Klingon).
+
+Changes in version 1.0 (major upgrade)
+ * Upgraded principal internal data structure from string to lattice.
+ * Improvements mostly in vowelization of South and Southeast Asian languages.
+ * Vocalic 'r' more consistently treated as vowel (no additional vowel added).
+ * Repetition signs (Japanese/Chinese/Thai/Khmer/Lao) are mapped to superscript 2.
+ * Japanese Katakana middle dots now mapped to ASCII space.
+ * Tibetan intersyllabic mark now mapped to middle dot (U+00B7).
+ * Some corrections regarding analysis of Chinese numbers.
+ * Many more foreign diacritics and punctuation marks dropped or mapped to ASCII.
+ * Zero-width characters dropped, except line/sentence-initial byte order marks.
+ * Spaces normalized to ASCII space.
+ * Fixed bug that in some cases mapped signs (such as dagger or bullet) to their verbal descriptions.
+ * Tested against previous version of uroman with a new uroman visual diff tool.
+ * Almost an order of magnitude faster.
+
+Changes in version 0.7 (minor upgrade)
+ * Added script uroman-quick.pl for Arabic script languages, incl. Uyghur.
+   Much faster, pre-caching mapping of Arabic to Latin characters, simple greedy processing.
+   Will not convert material from non-Arabic blocks such as any (somewhat unusual) Cyrillic
+   or Chinese characters in Uyghur texts.
+
+Changes in version 0.6 (minor upgrade)
+ * Added support for two letter characters used in Uzbek:
+   (1) character "ʻ" ("modifier letter turned comma", which modifies preceding "g" and "u" letters)
+   (2) character "ʼ" ("modifier letter apostrophe", which Uzbek uses to mark a glottal stop).
+   Both are now mapped to "'" (plain ASCII apostrophe).
+ * Added support for Uyghur vowel characters such as "ې" (Arabic e) and "ۆ" (Arabic oe)
+   even when they are not preceded by "ئ" (yeh with hamza above).
+ * Added support for Arabic semicolon "؛", Arabic ligature forms for phrases such as "ﷺ"
+   ("sallallahou alayhe wasallam" = "prayer of God be upon him and his family and peace")
+ * Added robustness for Arabic letter presentation forms (initial/medial/final/isolated).
+   However, it is strongly recommended to normalize any presentation form Arabic letters
+   to their non-presentation form before calling uroman.
+ * Added force flush directive ($|=1;).
+
+Changes in version 0.5 (minor upgrade)
+ * Improvements for Uyghur (make sure to use language option: -l uig)
+
+Changes in version 0.4 (minor upgrade)
+ * Improvements for Thai (special cases for vowel/consonant reordering, e.g. for "sara o"; dropped some aspiration 'h's)
+ * Minor change for Arabic (added "alef+fathatan" = "an")
+
+New features in version 0.3
+ * Covers Mandarin (Chinese)
+ * Improved romanization for numerous languages
+ * Preserves capitalization (e.g. from Latin, Cyrillic, Greek scripts)
+ * Maps from native digits to Western numbers
+ * Faster for South Asian languages
+
+### Other features
+ * Web interface: http://www.isi.edu/~ulf/uroman.html
+ * Vowelization is provided when locally computable, e.g. for many South Asian languages and Tibetan.
+
+### Limitations
+ * The current version of uroman has a few limitations, some of which we plan to address in future versions.
+   For Japanese, *uroman* currently romanizes hiragana and katakana as expected, but kanji are interpreted as Chinese characters and romanized as such. 
+   For Egyptian hieroglyphs, only single-sound phonetic characters and numbers are currently romanized. 
+   For Linear B, only phonetic syllabic characters are romanized. 
+   For some other extinct scripts such as cuneiform, no romanization is provided.
+ * A romanizer is not a full transliterator. For example, this version of
+   uroman does not vowelize text that lacks explicit vowelization such as
+   normal text in Arabic and Hebrew (without diacritics/points).
+
+### Acknowledgments
+This research is based upon work supported in part by the Office of the Director of National Intelligence (ODNI), Intelligence Advanced Research Projects Activity (IARPA), via contract # FA8650-17-C-9116, and by research sponsored by Air Force Research Laboratory (AFRL) under agreement number FA8750-19-1-1000. The views and conclusions contained herein are those of the authors and should not be interpreted as necessarily representing the official policies, either expressed or implied, of ODNI, IARPA, Air Force Laboratory, DARPA, or the U.S. Government. The U.S. Government is authorized to reproduce and distribute reprints for governmental purposes notwithstanding any copyright annotation therein.

pipeline/utils/uroman/README.txt

@@ -0,0 +1,141 @@
+uroman version 1.2.8
+Release date: April 23, 2021
+Author: Ulf Hermjakob, USC Information Sciences Institute
+
+uroman is a universal romanizer. It converts text in any script to the Latin alphabet.
+
+Usage: uroman.pl [-l <lang-code>] [--chart] [--no-cache] < STDIN
+       where the optional <lang-code> is a 3-letter languages code, e.g. ara, bel, bul, deu, ell, eng, fas,
+            grc, ell, eng, heb, kaz, kir, lav, lit, mkd, mkd2, oss, pnt, pus, rus, srp, srp2, tur, uig, ukr, yid.
+       --chart specifies chart output (in JSON format) to represent alternative romanizations.
+       --no-cache disables caching.
+Examples: bin/uroman.pl < text/zho.txt
+          bin/uroman.pl -l tur < text/tur.txt
+	  bin/uroman.pl -l heb --chart < text/heb.txt
+	  bin/uroman.pl < test/multi-script.txt > test/multi-script.uroman.txt
+
+Identifying the input as Arabic, Belarusian, Bulgarian, English, Farsi, German, 
+Ancient Greek, Modern Greek, Pontic Greek, Hebrew, Kazakh, Kyrgyz, Latvian, 
+Lithuanian, North Macedonian, Russian, Serbian, Turkish, Ukrainian, Uyghur or Yiddish 
+will improve romanization for those languages as some letters in those languages 
+have different sound values from other languages using the same script.
+No effect for other languages in this version.
+
+Bibliography: Ulf Hermjakob, Jonathan May, and Kevin Knight. 2018. Out-of-the-box universal romanization tool uroman. In Proceedings of the 56th Annual Meeting of Association for Computational Linguistics, Demo Track. [Best Demo Paper Award]
+
+Changes in version 1.2.8
+ * Improved support for Georgian.
+ * Updated UnicodeData.txt to version 13 (2021) with several new scripts (10% larger).
+ * Preserve various symbols (as opposed to mapping to the symbols' names).
+ * Various small improvements.
+Changes in version 1.2.7
+ * Improved support for Pashto.
+Changes in version 1.2.6
+ * Improved support for Ukrainian, Russian and Ogham (ancient Irish script).
+ * Added support for English Braille.
+ * Added alternative Romanization for North Macedonian and Serbian (mkd2/srp2)
+   reflecting a casual style that many native speakers of those languages use
+   when writing text in Latin script, e.g. non-accented single letters (e.g. "s")
+   rather than phonetically motivated combinations of letters (e.g. "sh").
+ * When a line starts with "::lcode xyz ", the new uroman version will switch to
+   that language for that line. This is used for the new reference test file.
+ * Various small improvements.
+Changes in version 1.2.5
+ * Improved support for Armenian and eight languages using Cyrillic scripts.
+   -- For Serbian and Macedonian, which are often written in both Cyrillic
+      and Latin scripts, uroman will map both official versions to the same
+      romanized text, e.g. both "Ниш" and "Niš" will be mapped to "Nish" (which
+      properly reflects the pronunciation of the city's name).
+      For both Serbian and Macedonian, casual writers often use a simplified
+      Latin form without diacritics, e.g. "s" to represent not only Cyrillic "с"
+      and Latin "s", but also "ш" or "š", even if this conflates "s" and "sh" and
+      other such pairs. The casual romanization can be simulated by using
+      alternative uroman language codes "srp2" and "mkd2", which romanize
+      both "Ниш" and "Niš" to "Nis" to reflect the casual Latin spelling.
+ * Various small improvements.
+Changes in version 1.2.4
+ * Added support for Tifinagh (a script used for Berber languages).
+ * Bug-fix that generated two emtpy lines for each empty line in cache mode.
+Changes in version 1.2.3
+ * Exclude emojis, dingbats, many other pictographs from being romanized (e.g. to "face")
+Changes in version 1.2
+ * Run-time improvement based on (1) token-based caching and (2) shortcut 
+   romanization (identity) of ASCII strings for default 1-best (non-chart) 
+   output. Speed-up by a factor of 10 for Bengali and Uyghur on medium and 
+   large size texts.
+ * Incremental improvements for Farsi, Amharic, Russian, Hebrew and related
+   languages.
+ * Richer lattice structure (more alternatives) for "Romanization" of English
+   to support better matching to romanizations of other languages.
+   Changes output only when --chart option is specified. No change in output for
+   default 1-best output, which for ASCII characters is always the input string.
+Changes in version 1.1 (major upgrade)
+ * Offers chart output (in JSON format) to represent alternative romanizations.
+   -- Location of first character is defined to be "line: 1, start:0, end:0".
+ * Incremental improvements of Hebrew and Greek romanization; Chinese numbers.
+ * Improved web-interface at http://www.isi.edu/~ulf/uroman.html
+   -- Shows corresponding original and romanization text in red
+      when hovering over a text segment.
+   -- Shows alternative romanizations when hovering over romanized text
+      marked by dotted underline.
+   -- Added right-to-left script detection and improved display for right-to-left
+      script text (as determined line by line).
+   -- On-page support for some scripts that are often not pre-installed on users'
+      computers (Burmese, Egyptian, Klingon).
+Changes in version 1.0 (major upgrade)
+ * Upgraded principal internal data structure from string to lattice.
+ * Improvements mostly in vowelization of South and Southeast Asian languages.
+ * Vocalic 'r' more consistently treated as vowel (no additional vowel added).
+ * Repetition signs (Japanese/Chinese/Thai/Khmer/Lao) are mapped to superscript 2.
+ * Japanese Katakana middle dots now mapped to ASCII space.
+ * Tibetan intersyllabic mark now mapped to middle dot (U+00B7).
+ * Some corrections regarding analysis of Chinese numbers.
+ * Many more foreign diacritics and punctuation marks dropped or mapped to ASCII.
+ * Zero-width characters dropped, except line/sentence-initial byte order marks.
+ * Spaces normalized to ASCII space.
+ * Fixed bug that in some cases mapped signs (such as dagger or bullet) to their verbal descriptions.
+ * Tested against previous version of uroman with a new uroman visual diff tool.
+ * Almost an order of magnitude faster.
+Changes in version 0.7 (minor upgrade)
+ * Added script uroman-quick.pl for Arabic script languages, incl. Uyghur.
+   Much faster, pre-caching mapping of Arabic to Latin characters, simple greedy processing.
+   Will not convert material from non-Arabic blocks such as any (somewhat unusual) Cyrillic
+   or Chinese characters in Uyghur texts.
+Changes in version 0.6 (minor upgrade)
+ * Added support for two letter characters used in Uzbek:
+   (1) character "ʻ" ("modifier letter turned comma", which modifies preceding "g" and "u" letters)
+   (2) character "ʼ" ("modifier letter apostrophe", which Uzbek uses to mark a glottal stop).
+   Both are now mapped to "'" (plain ASCII apostrophe).
+ * Added support for Uyghur vowel characters such as "ې" (Arabic e) and "ۆ" (Arabic oe)
+   even when they are not preceded by "ئ" (yeh with hamza above).
+ * Added support for Arabic semicolon "؛", Arabic ligature forms for phrases such as "ﷺ"
+   ("sallallahou alayhe wasallam" = "prayer of God be upon him and his family and peace")
+ * Added robustness for Arabic letter presentation forms (initial/medial/final/isolated).
+   However, it is strongly recommended to normalize any presentation form Arabic letters
+   to their non-presentation form before calling uroman.
+ * Added force flush directive ($|=1;).
+Changes in version 0.5 (minor upgrade)
+ * Improvements for Uyghur (make sure to use language option: -l uig)
+Changes in version 0.4 (minor upgrade)
+ * Improvements for Thai (special cases for vowel/consonant reordering, e.g. for "sara o"; dropped some aspiration 'h's)
+ * Minor change for Arabic (added "alef+fathatan" = "an")
+New features in version 0.3
+ * Covers Mandarin (Chinese)
+ * Improved romanization for numerous languages
+ * Preserves capitalization (e.g. from Latin, Cyrillic, Greek scripts)
+ * Maps from native digits to Western numbers
+ * Faster for South Asian languages
+
+Other features
+ * Web interface: http://www.isi.edu/~ulf/uroman.html
+ * Vowelization is provided when locally computable, e.g. for many South Asian
+   languages and Tibetan.
+
+Limitations
+ * This version of uroman assumes all CJK ideographs to be Mandarin (Chinese).
+   This means that Japanese kanji are incorrectly romanized; however, Japanese
+   hiragana and katakana are properly romanized.
+ * A romanizer is not a full transliterator. For example, this version of
+   uroman does not vowelize text that lacks explicit vowelization such as
+   normal text in Arabic and Hebrew (without diacritics/points).
+

pipeline/utils/uroman/bin/de-accent.pl

@@ -0,0 +1,201 @@
+#!/usr/bin/perl -w
+
+sub print_version {
+   print STDERR "$0 version 1.1\n";
+   print STDERR "   Author: Ulf Hermjakob\n";
+   print STDERR "   Last changed: March 14, 2011\n";
+}
+
+sub print_usage {
+   print STDERR "$0 [options] < with_accents.txt > without_accents.txt\n";
+   print STDERR "   -h or -help\n";
+   print STDERR "   -v or -version\n";
+}
+
+sub de_accent_string {
+   local($s) = @_;
+
+   # $s =~ tr/A-Z/a-z/;
+   unless (0) {
+      # Latin-1
+      if ($s =~ /\xC3[\x80-\xBF]/) {
+         $s =~ s/(À|Á|Â|Ã|Ä|Å)/A/g;
+         $s =~ s/Æ/Ae/g;
+         $s =~ s/Ç/C/g;
+         $s =~ s/Ð/D/g;
+         $s =~ s/(È|É|Ê|Ë)/E/g;
+         $s =~ s/(Ì|Í|Î|Ï)/I/g;
+         $s =~ s/Ñ/N/g;
+         $s =~ s/(Ò|Ó|Ô|Õ|Ö|Ø)/O/g;
+         $s =~ s/(Ù|Ú|Û|Ü)/U/g;
+         $s =~ s/Þ/Th/g;
+         $s =~ s/Ý/Y/g;
+         $s =~ s/(à|á|â|ã|ä|å)/a/g;
+         $s =~ s/æ/ae/g;
+         $s =~ s/ç/c/g;
+         $s =~ s/(è|é|ê|ë)/e/g;
+         $s =~ s/(ì|í|î|ï)/i/g;
+         $s =~ s/ð/d/g;
+         $s =~ s/ñ/n/g;
+         $s =~ s/(ò|ó|ô|õ|ö)/o/g;
+         $s =~ s/ß/ss/g;
+         $s =~ s/þ/th/g;
+         $s =~ s/(ù|ú|û|ü)/u/g;
+         $s =~ s/(ý|ÿ)/y/g;
+      }
+      # Latin Extended-A
+      if ($s =~ /[\xC4-\xC5][\x80-\xBF]/) {
+         $s =~ s/(Ā|Ă|Ą)/A/g;
+         $s =~ s/(ā|ă|ą)/a/g;
+         $s =~ s/(Ć|Ĉ|Ċ|Č)/C/g;
+         $s =~ s/(ć|ĉ|ċ|č)/c/g;
+         $s =~ s/(Ď|Đ)/D/g;
+         $s =~ s/(ď|đ)/d/g;
+         $s =~ s/(Ē|Ĕ|Ė|Ę|Ě)/E/g;
+         $s =~ s/(ē|ĕ|ė|ę|ě)/e/g;
+         $s =~ s/(Ĝ|Ğ|Ġ|Ģ)/G/g;
+         $s =~ s/(ĝ|ğ|ġ|ģ)/g/g;
+         $s =~ s/(Ĥ|Ħ)/H/g;
+         $s =~ s/(ĥ|ħ)/h/g;
+         $s =~ s/(Ĩ|Ī|Ĭ|Į|İ)/I/g;
+         $s =~ s/(ĩ|ī|ĭ|į|ı)/i/g;
+         $s =~ s/IJ/Ij/g;
+         $s =~ s/ij/ij/g;
+         $s =~ s/Ĵ/J/g;
+         $s =~ s/ĵ/j/g;
+         $s =~ s/Ķ/K/g;
+         $s =~ s/(ķ|ĸ)/k/g;
+         $s =~ s/(Ĺ|Ļ|Ľ|Ŀ|Ł)/L/g;
+         $s =~ s/(ļ|ľ|ŀ|ł)/l/g;
+         $s =~ s/(Ń|Ņ|Ň|Ŋ)/N/g;
+         $s =~ s/(ń|ņ|ň|ʼn|ŋ)/n/g;
+         $s =~ s/(Ō|Ŏ|Ő)/O/g;
+         $s =~ s/(ō|ŏ|ő)/o/g;
+         $s =~ s/Œ/Oe/g;
+         $s =~ s/œ/oe/g;
+         $s =~ s/(Ŕ|Ŗ|Ř)/R/g;
+         $s =~ s/(ŕ|ŗ|ř)/r/g;
+         $s =~ s/(Ś|Ŝ|Ş|Š)/S/g;
+         $s =~ s/(ś|ŝ|ş|š|ſ)/s/g;
+         $s =~ s/(Ţ|Ť|Ŧ)/T/g;
+         $s =~ s/(ţ|ť|ŧ)/t/g;
+         $s =~ s/(Ũ|Ū|Ŭ|Ů|Ű|Ų)/U/g;
+         $s =~ s/(ũ|ū|ŭ|ů|ű|ų)/u/g;
+         $s =~ s/Ŵ/W/g;
+         $s =~ s/ŵ/w/g;
+         $s =~ s/(Ŷ|Ÿ)/Y/g;
+         $s =~ s/ŷ/y/g;
+         $s =~ s/(Ź|Ż|Ž)/Z/g;
+         $s =~ s/(ź|ż|ž)/z/g;
+      }
+      # Latin Extended Additional
+      if ($s =~ /\xE1[\xB8-\xBF][\x80-\xBF]/) {
+          $s =~ s/(ḁ|ạ|ả|ấ|ầ|ẩ|ẫ|ậ|ắ|ằ|ẳ|ẵ|ặ|ẚ)/a/g;
+          $s =~ s/(ḃ|ḅ|ḇ)/b/g;
+          $s =~ s/(ḉ)/c/g;
+          $s =~ s/(ḋ|ḍ|ḏ|ḑ|ḓ)/d/g;
+          $s =~ s/(ḕ|ḗ|ḙ|ḛ|ḝ|ẹ|ẻ|ẽ|ế|ề|ể|ễ|ệ)/e/g;
+          $s =~ s/(ḟ)/f/g;
+          $s =~ s/(ḡ)/g/g;
+          $s =~ s/(ḣ|ḥ|ḧ|ḩ|ḫ)/h/g;
+          $s =~ s/(ḭ|ḯ|ỉ|ị)/i/g;
+          $s =~ s/(ḱ|ḳ|ḵ)/k/g;
+          $s =~ s/(ḷ|ḹ|ḻ|ḽ)/l/g;
+          $s =~ s/(ḿ|ṁ|ṃ)/m/g;
+          $s =~ s/(ṅ|ṇ|ṉ|ṋ)/m/g;
+          $s =~ s/(ọ|ỏ|ố|ồ|ổ|ỗ|ộ|ớ|ờ|ở|ỡ|ợ|ṍ|ṏ|ṑ|ṓ)/o/g;
+          $s =~ s/(ṕ|ṗ)/p/g;
+          $s =~ s/(ṙ|ṛ|ṝ|ṟ)/r/g;
+          $s =~ s/(ṡ|ṣ|ṥ|ṧ|ṩ|ẛ)/s/g;
+          $s =~ s/(ṫ|ṭ|ṯ|ṱ)/t/g;
+          $s =~ s/(ṳ|ṵ|ṷ|ṹ|ṻ|ụ|ủ|ứ|ừ|ử|ữ|ự)/u/g;
+          $s =~ s/(ṽ|ṿ)/v/g;
+          $s =~ s/(ẁ|ẃ|ẅ|ẇ|ẉ|ẘ)/w/g;
+          $s =~ s/(ẋ|ẍ)/x/g;
+          $s =~ s/(ẏ|ỳ|ỵ|ỷ|ỹ|ẙ)/y/g;
+          $s =~ s/(ẑ|ẓ|ẕ)/z/g;
+          $s =~ s/(Ḁ|Ạ|Ả|Ấ|Ầ|Ẩ|Ẫ|Ậ|Ắ|Ằ|Ẳ|Ẵ|Ặ)/A/g;
+          $s =~ s/(Ḃ|Ḅ|Ḇ)/B/g;
+          $s =~ s/(Ḉ)/C/g;
+          $s =~ s/(Ḋ|Ḍ|Ḏ|Ḑ|Ḓ)/D/g;
+          $s =~ s/(Ḕ|Ḗ|Ḙ|Ḛ|Ḝ|Ẹ|Ẻ|Ẽ|Ế|Ề|Ể|Ễ|Ệ)/E/g;
+          $s =~ s/(Ḟ)/F/g;
+          $s =~ s/(Ḡ)/G/g;
+          $s =~ s/(Ḣ|Ḥ|Ḧ|Ḩ|Ḫ)/H/g;
+          $s =~ s/(Ḭ|Ḯ|Ỉ|Ị)/I/g;
+          $s =~ s/(Ḱ|Ḳ|Ḵ)/K/g;
+          $s =~ s/(Ḷ|Ḹ|Ḻ|Ḽ)/L/g;
+          $s =~ s/(Ḿ|Ṁ|Ṃ)/M/g;
+          $s =~ s/(Ṅ|Ṇ|Ṉ|Ṋ)/N/g;
+          $s =~ s/(Ṍ|Ṏ|Ṑ|Ṓ|Ọ|Ỏ|Ố|Ồ|Ổ|Ỗ|Ộ|Ớ|Ờ|Ở|Ỡ|Ợ)/O/g;
+          $s =~ s/(Ṕ|Ṗ)/P/g;
+          $s =~ s/(Ṙ|Ṛ|Ṝ|Ṟ)/R/g;
+          $s =~ s/(Ṡ|Ṣ|Ṥ|Ṧ|Ṩ)/S/g;
+          $s =~ s/(Ṫ|Ṭ|Ṯ|Ṱ)/T/g;
+          $s =~ s/(Ṳ|Ṵ|Ṷ|Ṹ|Ṻ|Ụ|Ủ|Ứ|Ừ|Ử|Ữ|Ự)/U/g;
+          $s =~ s/(Ṽ|Ṿ)/V/g;
+          $s =~ s/(Ẁ|Ẃ|Ẅ|Ẇ|Ẉ)/W/g;
+          $s =~ s/(Ẍ)/X/g;
+          $s =~ s/(Ẏ|Ỳ|Ỵ|Ỷ|Ỹ)/Y/g; 
+          $s =~ s/(Ẑ|Ẓ|Ẕ)/Z/g;
+      }
+      # Greek letters
+      if ($s =~ /\xCE[\x86-\xAB]/) {
+          $s =~ s/ά/α/g;
+          $s =~ s/έ/ε/g;
+          $s =~ s/ί/ι/g;
+          $s =~ s/ϊ/ι/g;
+          $s =~ s/ΐ/ι/g;
+          $s =~ s/ό/ο/g;
+          $s =~ s/ύ/υ/g;
+          $s =~ s/ϋ/υ/g;
+          $s =~ s/ΰ/υ/g;
+          $s =~ s/ώ/ω/g;
+          $s =~ s/Ά/Α/g;
+          $s =~ s/Έ/Ε/g;
+          $s =~ s/Ή/Η/g;
+          $s =~ s/Ί/Ι/g;
+          $s =~ s/Ϊ/Ι/g;
+          $s =~ s/Ύ/Υ/g;
+          $s =~ s/Ϋ/Υ/g;
+          $s =~ s/Ώ/Ω/g;
+      }
+      # Cyrillic letters
+      if ($s =~ /\xD0[\x80-\xAF]/) {
+          $s =~ s/Ѐ/Е/g;
+          $s =~ s/Ё/Е/g;
+          $s =~ s/Ѓ/Г/g;
+          $s =~ s/Ќ/К/g;
+          $s =~ s/Ѝ/И/g;
+          $s =~ s/Й/И/g;
+          $s =~ s/ѐ/е/g;
+          $s =~ s/ё/е/g;
+          $s =~ s/ѓ/г/g;
+          $s =~ s/ќ/к/g;
+          $s =~ s/ѝ/и/g;
+          $s =~ s/й/и/g;
+      }
+   }
+   return $s;
+}
+
+while (@ARGV) {
+   $arg = shift @ARGV;
+   if ($arg =~ /^-*(h|help)$/i) {
+      &print_usage;
+      exit 1;
+   } elsif ($arg =~ /^-*(v|version)$/i) {
+      &print_version;
+      exit 1;
+   } else {
+      print STDERR "Ignoring unrecognized argument $arg\n";
+   }
+}
+
+$line_number = 0;
+while (<>) {
+   $line_number++;
+   print &de_accent_string($_);
+}
+exit 0;
+

pipeline/utils/uroman/bin/string-distance.pl

@@ -0,0 +1,99 @@
+#!/usr/bin/perl -w
+
+# Author: Ulf Hermjakob
+# Release date: October 13, 2019
+
+# Usage: string-distance.pl {-lc1 <language-code>} {-lc2 <language-code>} < STDIN > STDOUT
+# Example: string-distance.pl -lc1 rus -lc2 ukr < STDIN > STDOUT
+# Example: string-distance.pl < ../test/string-similarity-test-input.txt
+# Input format: two strings per line (tab-separated, in Latin script)
+#    Strings in non-Latin scripts should first be romanized. (Recommended script: uroman.pl)
+# Output format: repetition of the two input strings, plus the string distance between them (tab-separated).
+#    Additional output meta info lines at the top are marked with an initial #.
+#
+# The script uses data from a string-distance-cost-rules file that lists costs,
+# where the default cost is "1" with lower costs for differences in vowels,
+# duplicate consonants, "f" vs. "ph" etc.
+# Language cost rules can be language-specific and context-sensitive.
+
+$|=1;
+
+use FindBin;
+use Cwd "abs_path";
+use File::Basename qw(dirname);
+use File::Spec;
+
+my $bin_dir = abs_path(dirname($0));
+my $root_dir = File::Spec->catfile($bin_dir, File::Spec->updir());
+my $data_dir = File::Spec->catfile($root_dir, "data");
+my $lib_dir = File::Spec->catfile($root_dir, "lib");
+
+use lib "$FindBin::Bin/../lib";
+use List::Util qw(min max);
+use NLP::utilities;
+use NLP::stringDistance;
+$util = NLP::utilities;
+$sd = NLP::stringDistance;
+$verbose = 0;
+$separator = "\t";
+
+$cost_rule_filename = File::Spec->catfile($data_dir, "string-distance-cost-rules.txt");
+
+$lang_code1 = "eng";
+$lang_code2 = "eng";
+%ht = ();
+
+while (@ARGV) {
+   $arg = shift @ARGV;
+   if ($arg =~ /^-+lc1$/) {
+      $lang_code_candidate = shift @ARGV;
+      $lang_code1 = $lang_code_candidate if $lang_code_candidate =~ /^[a-z]{3,3}$/;
+   } elsif ($arg =~ /^-+lc2$/) {
+      $lang_code_candidate = shift @ARGV;
+      $lang_code2 = $lang_code_candidate if $lang_code_candidate =~ /^[a-z]{3,3}$/;
+   } elsif ($arg =~ /^-+(v|verbose)$/) {
+      $verbose = shift @ARGV;
+   } else {
+      print STDERR "Ignoring unrecognized arg $arg\n";
+   }
+}
+
+$sd->load_string_distance_data($cost_rule_filename, *ht, $verbose);
+print STDERR "Loaded resources.\n" if $verbose;
+
+my $chart_id = 0;
+my $line_number = 0;
+print "# Lang-code-1: $lang_code1 Lang-code-2: $lang_code2\n";
+while (<>) {
+   $line_number++;
+   if ($verbose) {
+      if ($line_number =~ /000$/) {
+         if ($line_number =~ /0000$/) {
+	    print STDERR $line_number;
+         } else {
+	    print STDERR ".";
+         }
+      }
+   }
+   my $line = $_;
+   $line =~ s/^\xEF\xBB\xBF//;
+   next if $line =~ /^\s*(\#.*)?$/;
+   my $s1;
+   my $s2;
+   if (($s1, $s2) = ($line =~ /^("(?:\\"|[^"])*"|\S+)$separator("(?:\\"|[^"])*"|\S+)\s*$/)) {
+      $s1 = $util->dequote_string($s1);
+      $s2 = $util->dequote_string($s2);
+   } elsif ($line =~ /^\s*(#.*)$/) {
+   } else {
+      print STDERR "Could not process line $line_number: $line" if $verbose;
+      print "\n";
+      next;
+   }
+
+   $cost = $sd->quick_romanized_string_distance_by_chart($s1, $s2, *ht, "", $lang_code1, $lang_code2);
+   print "$s1\t$s2\t$cost\n";
+}
+print STDERR "\n" if $verbose;
+
+exit 0;
+

pipeline/utils/uroman/bin/uroman-quick.pl

@@ -0,0 +1,58 @@
+#!/usr/bin/perl -w
+
+# uroman  Nov. 12, 2015 - July 25, 2016
+# version v0.7
+# Author: Ulf Hermjakob
+
+# Usage: uroman-quick.pl {-l [tur|uig|ukr|yid]} < STDIN
+# currently only for Arabic script languages, incl. Uyghur
+
+$|=1;
+
+use FindBin;
+use Cwd "abs_path";
+use File::Basename qw(dirname);
+use File::Spec;
+
+my $bin_dir = abs_path(dirname($0));
+my $root_dir = File::Spec->catfile($bin_dir, File::Spec->updir());
+my $data_dir = File::Spec->catfile($root_dir, "data");
+my $lib_dir = File::Spec->catfile($root_dir, "lib");
+
+use lib "$FindBin::Bin/../lib";
+use NLP::Romanizer;
+use NLP::UTF8;
+$romanizer = NLP::Romanizer;
+%ht = ();
+$lang_code = "";
+
+while (@ARGV) {
+   $arg = shift @ARGV;
+   if ($arg =~ /^-+(l|lc|lang-code)$/) {
+      $lang_code = lc (shift @ARGV || "")
+   } else {
+      print STDERR "Ignoring unrecognized arg $arg\n";
+   }
+}
+
+$romanization_table_arabic_block_filename = File::Spec->catfile($data_dir, "romanization-table-arabic-block.txt");
+$romanization_table_filename = File::Spec->catfile($data_dir, "romanization-table.txt");
+
+$romanizer->load_romanization_table(*ht, $romanization_table_arabic_block_filename);
+$romanizer->load_romanization_table(*ht, $romanization_table_filename);
+
+$line_number = 0;
+while (<>) {
+   $line_number++;
+   my $line = $_;
+   print $romanizer->quick_romanize($line, $lang_code, *ht) . "\n";
+   if ($line_number =~ /0000$/) {
+      print STDERR $line_number;
+   } elsif ($line_number =~ /000$/) {
+      print STDERR ".";
+   }
+}
+print STDERR "\n";
+
+exit 0;
+

pipeline/utils/uroman/bin/uroman-tsv.sh

@@ -0,0 +1,28 @@
+#!/usr/bin/env bash
+# Created by Thamme Gowda on June 17, 2019
+
+DIR=$(dirname "${BASH_SOURCE[0]}")  # get the directory name
+# DIR=$(realpath "${DIR}")    # resolve its full path if need be
+
+if [[ $# -lt 1 || $# -gt 2 ]]; then
+    >&2 echo "ERROR: invalid args"
+    >&2 echo "Usage: <input.tsv> [<output.tsv>]"
+    exit 2
+fi
+
+INP=$1
+OUT=$2
+
+CMD=$DIR/uroman.pl
+
+function romanize(){
+    paste <(cut -f1 $INP) <(cut -f2 $INP | $CMD)
+}
+
+if [[ -n $OUT ]]; then
+    romanize > $OUT
+else
+    romanize
+fi
+
+

pipeline/utils/uroman/bin/uroman.pl

@@ -0,0 +1,138 @@
+#!/usr/bin/perl -w
+
+# uroman  Nov. 12, 2015 - Apr. 23, 2021
+$version = "v1.2.8";
+# Author: Ulf Hermjakob
+
+# Usage: uroman.pl {-l [ara|bel|bul|deu|ell|eng|fas|grc|heb|kaz|kir|lav|lit|mkd|mkd2|oss|pnt|rus|srp|srp2|tur|uig|ukr|yid]} {--chart|--offset-mapping} {--no-cache} {--workset} < STDIN
+# Example: cat workset.txt | uroman.pl --offset-mapping --workset
+
+$|=1;
+
+use FindBin;
+use Cwd "abs_path";
+use File::Basename qw(dirname);
+use File::Spec;
+
+my $bin_dir = abs_path(dirname($0));
+my $root_dir = File::Spec->catfile($bin_dir, File::Spec->updir());
+my $data_dir = File::Spec->catfile($root_dir, "data");
+my $lib_dir = File::Spec->catfile($root_dir, "lib");
+
+use lib "$FindBin::Bin/../lib";
+use NLP::Chinese;
+use NLP::Romanizer;
+use NLP::UTF8;
+use NLP::utilities;
+use JSON;
+$chinesePM = NLP::Chinese;
+$romanizer = NLP::Romanizer;
+$util = NLP::utilities;
+%ht = ();
+%pinyin_ht = ();
+$lang_code = "";
+$return_chart_p = 0;
+$return_offset_mappings_p = 0;
+$workset_p = 0;
+$cache_rom_tokens_p = 1;
+
+$script_data_filename = File::Spec->catfile($data_dir, "Scripts.txt");
+$unicode_data_overwrite_filename = File::Spec->catfile($data_dir, "UnicodeDataOverwrite.txt");
+$unicode_data_filename = File::Spec->catfile($data_dir, "UnicodeData.txt");
+$romanization_table_filename = File::Spec->catfile($data_dir, "romanization-table.txt");
+$chinese_tonal_pinyin_filename = File::Spec->catfile($data_dir, "Chinese_to_Pinyin.txt");
+
+while (@ARGV) {
+   $arg = shift @ARGV;
+   if ($arg =~ /^-+(l|lc|lang-code)$/) {
+      $lang_code = lc (shift @ARGV || "")
+   } elsif ($arg =~ /^-+chart$/i) {
+      $return_chart_p = 1;
+   } elsif ($arg =~ /^-+workset$/i) {
+      $workset_p = 1;
+   } elsif ($arg =~ /^-+offset[-_]*map/i) {
+      $return_offset_mappings_p = 1;
+   } elsif ($arg =~ /^-+unicode[-_]?data/i) {
+      $filename = shift @ARGV;
+      if (-r $filename) {
+         $unicode_data_filename = $filename;
+      } else {
+         print STDERR "Ignoring invalid UnicodeData filename $filename\n";
+      }
+   } elsif ($arg =~ /^-+(no-tok-cach|no-cach)/i) {
+      $cache_rom_tokens_p = 0;
+   } else {
+      print STDERR "Ignoring unrecognized arg $arg\n";
+   }
+}
+
+$romanizer->load_script_data(*ht, $script_data_filename);
+$romanizer->load_unicode_data(*ht, $unicode_data_filename);
+$romanizer->load_unicode_overwrite_romanization(*ht, $unicode_data_overwrite_filename);
+$romanizer->load_romanization_table(*ht, $romanization_table_filename);
+$chinese_to_pinyin_not_yet_loaded_p = 1;
+$current_date = $util->datetime("dateTtime");
+$lang_code_clause = ($lang_code) ? " \"lang-code\":\"$lang_code\",\n" : "";
+
+print "{\n \"romanizer\":\"uroman $version (Ulf Hermjakob, USC/ISI)\",\n \"date\":\"$current_date\",\n$lang_code_clause \"romanization\": [\n" if $return_chart_p;
+my $line_number = 0;
+my $chart_result = "";
+while (<>) {
+   $line_number++;
+   my $line = $_;
+   my $snt_id = "";
+   if ($workset_p) {
+      next if $line =~ /^#/;
+      if (($i_value, $s_value) = ($line =~ /^(\S+\.\d+)\s(.*)$/)) {
+	 $snt_id = $i_value;
+	 $line = "$s_value\n";
+      } else {
+	 next;
+      }
+   }
+   if ($chinese_to_pinyin_not_yet_loaded_p && $chinesePM->string_contains_utf8_cjk_unified_ideograph_p($line)) {
+      $chinesePM->read_chinese_tonal_pinyin_files(*pinyin_ht, $chinese_tonal_pinyin_filename);
+      $chinese_to_pinyin_not_yet_loaded_p = 0;
+   }
+   if ($return_chart_p) {
+      print $chart_result;
+      *chart_ht = $romanizer->romanize($line, $lang_code, "", *ht, *pinyin_ht, 0, "return chart", $line_number);
+      $chart_result = $romanizer->chart_to_json_romanization_elements(0, $chart_ht{N_CHARS}, *chart_ht, $line_number);
+   } elsif ($return_offset_mappings_p) {
+      ($best_romanization, $offset_mappings) = $romanizer->romanize($line, $lang_code, "", *ht, *pinyin_ht, 0, "return offset mappings", $line_number, 0);
+      print "::snt-id $snt_id\n" if $workset_p;
+      print "::orig $line";
+      print "::rom $best_romanization\n";
+      print "::align $offset_mappings\n\n";
+   } elsif ($cache_rom_tokens_p) {
+      print $romanizer->romanize_by_token_with_caching($line, $lang_code, "", *ht, *pinyin_ht, 0, "", $line_number) . "\n";
+   } else {
+      print $romanizer->romanize($line, $lang_code, "", *ht, *pinyin_ht, 0, "", $line_number) . "\n";
+   }
+}
+$chart_result =~ s/,(\s*)$/$1/;
+print $chart_result;
+print " ]\n}\n" if $return_chart_p;
+
+$dev_test_p = 0;
+if ($dev_test_p) {
+   $n_suspicious_code_points = 0;
+   $n_instances = 0;
+   foreach $char_name (sort { hex($ht{UTF_NAME_TO_UNICODE}->{$a}) <=> hex($ht{UTF_NAME_TO_UNICODE}->{$b}) }
+                            keys %{$ht{SUSPICIOUS_ROMANIZATION}}) {
+      $unicode_value = $ht{UTF_NAME_TO_UNICODE}->{$char_name};
+      $utf8_string = $ht{UTF_NAME_TO_CODE}->{$char_name};
+      foreach $romanization (sort keys %{$ht{SUSPICIOUS_ROMANIZATION}->{$char_name}}) {
+         $count = $ht{SUSPICIOUS_ROMANIZATION}->{$char_name}->{$romanization};
+	 $s = ($count == 1) ? "" : "s";
+         print STDERR "*** Suspiciously lengthy romanization:\n" unless $n_suspicious_code_points;
+         print STDERR "::s $utf8_string ::t $romanization ::comment $char_name (U+$unicode_value)\n";
+	 $n_suspicious_code_points++;
+	 $n_instances += $count;
+      }
+   }
+   print STDERR "  *** Total of $n_suspicious_code_points suspicious code points ($n_instances instance$s)\n" if $n_suspicious_code_points;
+} 
+
+exit 0;
+

pipeline/utils/uroman/data/Scripts.txt

@@ -0,0 +1,174 @@
+::script-name Adlam
+::script-name Aegean
+::script-name Ahom
+::script-name Anatolian Hieroglyph
+::script-name Arabic ::direction right-to-left
+::script-name Arabic-Indic
+::script-name Armenian
+::script-name Avestan
+::script-name Balinese
+::script-name Bamum
+::script-name Bassa Vah
+::script-name Batak
+::script-name Bengali ::abugida-default-vowel a
+::script-name Bhaiksuki
+::script-name Bopomofo ::language Chinese
+::script-name Brahmi ::abugida-default-vowel a
+::script-name Braille
+::script-name Buginese
+::script-name Buhid
+::script-name Canadian Syllabics
+::script-name Carian
+::script-name Caucasian Albanian
+::script-name Chakma
+::script-name Cham
+::script-name Cherokee
+::script-name Chorasmian
+::script-name Coptic
+::script-name Cuneiform
+::script-name Cypro-Minoan
+::script-name Cypriot
+::script-name Cyrillic
+::script-name CJK ::alt-script-name Chinese, Kanji ::language Chinese, Japanese, Korean, Mandarin
+::script-name Deseret
+::script-name Devanagari ::abugida-default-vowel a
+::script-name Dives Akuru
+::script-name Dogra
+::script-name Duployan
+::script-name Egyptian Hieroglyph ::alt-script-name Egyptian
+::script-name Elbasan
+::script-name Elymaic
+::script-name Ethiopic
+::script-name Extended Arabic-Indic
+::script-name Georgian
+::script-name Glagolitic
+::script-name Gothic
+::script-name Grantha
+::script-name Greek
+::script-name Greek Acrophonic
+::script-name Gujarati ::abugida-default-vowel a
+::script-name Gunjala Gondi
+::script-name Gurmukhi ::abugida-default-vowel a
+::script-name Hangul ::language Korean
+::script-name Hangzhou
+::script-name Hanifi Rohingya
+::script-name Hanunoo
+::script-name Hatran
+::script-name Hebrew ::direction right-to-left
+::script-name Hiragana ::language Japanese
+::script-name Indic Siyaq
+::script-name Imperial Aramaic
+::script-name Inscriptional Pahlavi
+::script-name Inscriptional Parthian
+::script-name Javanese
+::script-name Kaithi
+::script-name Kannada ::abugida-default-vowel a
+::script-name Katakana ::language Japanese
+::script-name Kawi
+::script-name Kayah Li
+::script-name Kharoshthi
+::script-name Khitan Small Script
+::script-name Khmer ::abugida-default-vowel a, o
+::script-name Khojki
+::script-name Khudawadi
+::script-name Klingon
+::script-name Lao
+::script-name Lepcha
+::script-name Latin
+::script-name Limbu
+::script-name Linear A
+::script-name Linear B
+::script-name Lisu
+::script-name Lycian
+::script-name Lydian
+::script-name Mahajani
+::script-name Makasar
+::script-name Malayalam ::abugida-default-vowel a
+::script-name Mandaic
+::script-name Manichaean
+::script-name Marchen
+::script-name Masaram Gondi
+::script-name Mayan
+::script-name Medefaidrin
+::script-name Meetei Mayek
+::script-name Mende Kikakui
+::script-name Meroitic Cursive
+::script-name Meroitic Hieroglyphic
+::script-name Miao
+::script-name Modi ::abugida-default-vowel a
+::script-name Mongolian
+::script-name Mro
+::script-name Multani
+::script-name Myanmar ::alt-script-name Burmese ::abugida-default-vowel a
+::script-name Nabataean
+::script-name Nag Mundari
+::script-name Nandinagari
+::script-name New Tai Lue
+::script-name Newa
+::script-name Nko ::direction right-to-left
+::script-name North Indic
+::script-name Nushu
+::script-name Nyiakeng Puachue Hmong
+::script-name Ogham
+::script-name Ol Chiki
+::script-name Old Hungarian
+::script-name Old Italic
+::script-name Old Permic
+::script-name Old Persian
+::script-name Old North Arabian
+::script-name Old Sogdian
+::script-name Old South Arabian
+::script-name Old Turkic
+::script-name Old Uyghur
+::script-name Oriya ::alt-script-name Odia ::abugida-default-vowel a
+::script-name Osage
+::script-name Osmanya
+::script-name Ottoman Siyaq
+::script-name Pahawh Hmong
+::script-name Palmyrene
+::script-name Pau Cin Hau
+::script-name Phags-Pa
+::script-name Phaistos Disc
+::script-name Phoenician
+::script-name Psalter Pahlavi
+::script-name Rejang
+::script-name Rumi
+::script-name Runic
+::script-name Samaritan
+::script-name Saurashtra
+::script-name Sharada
+::script-name Shavian
+::script-name Siddham
+::script-name SignWriting
+::script-name Sinhala ::abugida-default-vowel a
+::script-name Sogdian
+::script-name Sora Sompeng
+::script-name Soyombo
+::script-name Sundanese ::abugida-default-vowel a
+::script-name Syloti Nagri
+::script-name Syriac
+::script-name Tagalog
+::script-name Tagbanwa
+::script-name Tai Le
+::script-name Tai Tham
+::script-name Tai Viet
+::script-name Takri
+::script-name Tamil ::abugida-default-vowel a
+::script-name Tangsa
+::script-name Tangut
+::script-name Telugu ::abugida-default-vowel a
+::script-name Thaana ::direction right-to-left
+::script-name Thai
+::script-name Tibetan ::abugida-default-vowel a
+::script-name Tifinagh
+::script-name Tirhuta
+::script-name Toto
+::script-name Ugaritic
+::script-name Vai
+::script-name Vedic
+::script-name Vithkuqi
+::script-name Wancho
+::script-name Warang Citi
+::script-name Yezidi
+::script-name Yi
+::script-name Zanabazar Square