upload

README.md

@@ -0,0 +1,130 @@
+---
+language: ja
+datasets:
+- common_voice
+metrics:
+- wer
+tags:
+- audio
+- automatic-speech-recognition
+- speech
+- xlsr-fine-tuning-week
+license: apache-2.0
+model-index:
+- name: XLSR Wav2Vec2 Japanese Hiragana by Chien Vu
+  results:
+  - task:
+      name: Speech Recognition
+      type: automatic-speech-recognition
+    dataset:
+      name: Common Voice Japanese
+      type: common_voice
+      args: ja
+    metrics:
+       - name: Test WER
+         type: wer
+         value: 24.74
+       - name: Test CER
+         type: cer
+         value: 10.99
+---
+# Wav2Vec2-Large-XLSR-53-Japanese
+Fine-tuned [facebook/wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53) on Japanese using the [Common Voice](https://huggingface.co/datasets/common_voice) and Japanese speech corpus of Saruwatari-lab, University of Tokyo [JSUT](https://sites.google.com/site/shinnosuketakamichi/publication/jsut).
+When using this model, make sure that your speech input is sampled at 16kHz.
+## Usage
+The model can be used directly (without a language model) as follows:
+```python
+!pip install mecab-python3
+!pip install unidic-lite
+!pip install pykakasi
+!python -m unidic download
+import torch
+import torchaudio
+import librosa
+from datasets import load_dataset
+import MeCab
+from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
+import re
+# config
+wakati = MeCab.Tagger("-Owakati")
+chars_to_ignore_regex = '[\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\,\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\、\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\。\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\．\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\「\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\」\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\…\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\？\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\・]'
+kakasi = pykakasi.kakasi()
+kakasi.setMode("J","H")
+kakasi.setMode("K","H")
+kakasi.setMode("r","Hepburn")
+conv = kakasi.getConverter()
+# load data, processor and model
+test_dataset = load_dataset("common_voice", "ja", split="test[:2%]")
+processor = Wav2Vec2Processor.from_pretrained("vumichien/wav2vec2-large-xlsr-japanese-hỉragana")
+model = Wav2Vec2ForCTC.from_pretrained("vumichien/wav2vec2-large-xlsr-japanese-hỉragana")
+resampler = lambda sr, y: librosa.resample(y.numpy().squeeze(), sr, 16_000)
+# Preprocessing the datasets.
+def speech_file_to_array_fn(batch):
+    batch["sentence"] = conv.do(wakati.parse(batch["sentence"]).strip())
+    batch["sentence"] = re.sub(chars_to_ignore_regex,'', batch["sentence"]).strip()
+    speech_array, sampling_rate = torchaudio.load(batch["path"])
+    batch["speech"] = resampler(sampling_rate, speech_array).squeeze()
+    return batch
+test_dataset = test_dataset.map(speech_file_to_array_fn)
+inputs = processor(test_dataset["speech"][:2], sampling_rate=16_000, return_tensors="pt", padding=True)
+with torch.no_grad():
+    logits = model(inputs.input_values, attention_mask=inputs.attention_mask).logits
+predicted_ids = torch.argmax(logits, dim=-1)
+print("Prediction:", processor.batch_decode(predicted_ids))
+print("Reference:", test_dataset["sentence"][:2])
+```
+## Evaluation
+The model can be evaluated as follows on the Japanese test data of Common Voice.
+```python
+!pip install mecab-python3
+!pip install unidic-lite
+!pip install pykakasi
+!python -m unidic download
+import torch
+import librosa
+import torchaudio
+from datasets import load_dataset, load_metric
+import MeCab
+from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
+import re
+#config
+wakati = MeCab.Tagger("-Owakati")
+chars_to_ignore_regex = '[\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\,\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\、\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\。\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\．\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\「\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\」\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\…\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\？\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\・]'
+kakasi = pykakasi.kakasi()
+kakasi.setMode("J","H")
+kakasi.setMode("K","H")
+kakasi.setMode("r","Hepburn")
+conv = kakasi.getConverter()
+# load data, processor and model
+test_dataset = load_dataset("common_voice", "ja", split="test")
+wer = load_metric("wer")
+cer = load_metric("cer")
+processor = Wav2Vec2Processor.from_pretrained("vumichien/wav2vec2-large-xlsr-japanese-hỉragana")
+model = Wav2Vec2ForCTC.from_pretrained("vumichien/wav2vec2-large-xlsr-japanese-hỉragana")
+model.to("cuda")
+resampler = lambda sr, y: librosa.resample(y.numpy().squeeze(), sr, 16_000)
+# Preprocessing the datasets.
+def speech_file_to_array_fn(batch):
+    batch["sentence"] = conv.do(wakati.parse(batch["sentence"]).strip())
+    batch["sentence"] = re.sub(chars_to_ignore_regex,'', batch["sentence"]).strip()
+    speech_array, sampling_rate = torchaudio.load(batch["path"])
+    batch["speech"] = resampler(sampling_rate, speech_array).squeeze()
+    return batch
+test_dataset = test_dataset.map(speech_file_to_array_fn)
+# evaluate function
+def evaluate(batch):
+    inputs = processor(batch["speech"], sampling_rate=16_000, return_tensors="pt", padding=True)
+    with torch.no_grad():
+        logits = model(inputs.input_values.to("cuda"), attention_mask=inputs.attention_mask.to("cuda")).logits
+    pred_ids = torch.argmax(logits, dim=-1)
+    batch["pred_strings"] = processor.batch_decode(pred_ids)
+    return batch
+result = test_dataset.map(evaluate, batched=True, batch_size=8)
+print("WER: {:2f}".format(100 * wer.compute(predictions=result["pred_strings"], references=result["sentence"])))
+print("CER: {:2f}".format(100 * cer.compute(predictions=result["pred_strings"], references=result["sentence"])))
+```
+## Test Result
+**WER:** 24.74%,
+**CER:** 10.99%
+## Training
+The Common Voice `train`, `validation` datasets and Japanese speech corpus datasets were used for training.

config.json

@@ -0,0 +1,76 @@
+{
+  "_name_or_path": "facebook/wav2vec2-large-xlsr-53",
+  "activation_dropout": 0.0,
+  "apply_spec_augment": true,
+  "architectures": [
+    "Wav2Vec2ForCTC"
+  ],
+  "attention_dropout": 0.15,
+  "bos_token_id": 1,
+  "conv_bias": true,
+  "conv_dim": [
+    512,
+    512,
+    512,
+    512,
+    512,
+    512,
+    512
+  ],
+  "conv_kernel": [
+    10,
+    3,
+    3,
+    3,
+    3,
+    2,
+    2
+  ],
+  "conv_stride": [
+    5,
+    2,
+    2,
+    2,
+    2,
+    2,
+    2
+  ],
+  "ctc_loss_reduction": "mean",
+  "ctc_zero_infinity": false,
+  "do_stable_layer_norm": true,
+  "eos_token_id": 2,
+  "feat_extract_activation": "gelu",
+  "feat_extract_dropout": 0.0,
+  "feat_extract_norm": "layer",
+  "feat_proj_dropout": 0.2,
+  "final_dropout": 0.0,
+  "gradient_checkpointing": true,
+  "hidden_act": "gelu",
+  "hidden_dropout": 0.15,
+  "hidden_size": 1024,
+  "initializer_range": 0.02,
+  "intermediate_size": 4096,
+  "layer_norm_eps": 1e-05,
+  "layerdrop": 0.15,
+  "mask_channel_length": 10,
+  "mask_channel_min_space": 1,
+  "mask_channel_other": 0.0,
+  "mask_channel_prob": 0.0,
+  "mask_channel_selection": "static",
+  "mask_feature_length": 10,
+  "mask_feature_prob": 0.0,
+  "mask_time_length": 10,
+  "mask_time_min_space": 1,
+  "mask_time_other": 0.0,
+  "mask_time_prob": 0.1,
+  "mask_time_selection": "static",
+  "model_type": "wav2vec2",
+  "num_attention_heads": 16,
+  "num_conv_pos_embedding_groups": 16,
+  "num_conv_pos_embeddings": 128,
+  "num_feat_extract_layers": 7,
+  "num_hidden_layers": 24,
+  "pad_token_id": 85,
+  "transformers_version": "4.5.0.dev0",
+  "vocab_size": 86
+}

preprocessor_config.json

@@ -0,0 +1,8 @@
+{
+  "do_normalize": true,
+  "feature_size": 1,
+  "padding_side": "right",
+  "padding_value": 0.0,
+  "return_attention_mask": true,
+  "sampling_rate": 16000
+}

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7018fae264b57c8db15f8a53d4f042270b10fdcc6014e6af8d87266614b310ed
+size 1262286423

special_tokens_map.json

@@ -0,0 +1 @@
+{"bos_token": "<s>", "eos_token": "</s>", "unk_token": "[UNK]", "pad_token": "[PAD]"}

tokenizer_config.json

@@ -0,0 +1 @@
+{"unk_token": "[UNK]", "bos_token": "<s>", "eos_token": "</s>", "pad_token": "[PAD]", "do_lower_case": false, "word_delimiter_token": "|"}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7071f91984a92d8f1940fd926ca4fa345dc79c3acf3e19064f276ceea9f3b620
+size 2415

vocab.json

@@ -0,0 +1 @@
+{"も": 0, "ぜ": 1, "ざ": 2, "っ": 3, "ゃ": 4, "ね": 5, "に": 6, "ば": 7, "ひ": 8, "し": 9, "ぼ": 10, "ち": 11, "ふ": 12, "な": 13, "ぃ": 14, "れ": 15, "づ": 16, "ぉ": 17, "ぺ": 18, "で": 19, "を": 20, "ぐ": 21, "が": 22, "き": 23, "る": 24, "ぴ": 25, "べ": 26, "ゔ": 27, "は": 28, "ゅ": 29, "う": 30, "ぱ": 31, "お": 32, "そ": 33, "ぢ": 34, "ど": 35, "ろ": 36, "ぬ": 37, "ま": 38, "あ": 39, "ほ": 40, "ず": 41, "み": 42, "て": 43, "く": 44, "ぶ": 45, "げ": 46, "ぁ": 47, "ゆ": 48, "こ": 49, "え": 50, "め": 51, "ご": 52, "け": 53, "さ": 54, "ぞ": 55, "と": 56, "ぽ": 57, "た": 58, "ゖ": 59, "び": 60, "す": 61, "む": 62, "へ": 63, "い": 64, "だ": 65, "つ": 66, "ぅ": 67, "ぇ": 68, "じ": 69, "ー": 70, "ぎ": 71, "か": 72, "ら": 73, "り": 74, "せ": 75, "ょ": 76, "の": 77, "や": 78, "ぷ": 80, "わ": 81, "ん": 82, "よ": 83, "|": 79, "[UNK]": 84, "[PAD]": 85}