add test waves.

app.py

@@ -27,6 +27,7 @@ from datetime import datetime
 import gradio as gr
 import torchaudio
 
+from examples import examples
 from model import get_pretrained_model, language_to_models, sample_rate
 
 languages = list(language_to_models.keys())
@@ -51,11 +52,11 @@ def build_html_output(s: str, style: str = "result_item_success"):
 
 
 def process_uploaded_file(
-    in_filename: str,
     language: str,
     repo_id: str,
     decoding_method: str,
     num_active_paths: int,
+    in_filename: str,
 ):
     if in_filename is None or in_filename == "":
         return "", build_html_output(
@@ -79,11 +80,11 @@ def process_uploaded_file(
 
 
 def process_microphone(
-    in_filename: str,
     language: str,
     repo_id: str,
     decoding_method: str,
     num_active_paths: int,
+    in_filename: str,
 ):
     if in_filename is None or in_filename == "":
         return "", build_html_output(
@@ -108,17 +109,17 @@ def process_microphone(
 
 
 def process(
-    in_filename: str,
     language: str,
     repo_id: str,
     decoding_method: str,
     num_active_paths: int,
+    in_filename: str,
 ):
-    logging.info(f"in_filename: {in_filename}")
     logging.info(f"language: {language}")
     logging.info(f"repo_id: {repo_id}")
     logging.info(f"decoding_method: {decoding_method}")
     logging.info(f"num_active_paths: {num_active_paths}")
+    logging.info(f"in_filename: {in_filename}")
 
     filename = convert_to_wav(in_filename)
 
@@ -159,8 +160,14 @@ def process(
     info = f"""
     Wave duration  : {duration: .3f} s <br/>
     Processing time: {end - start: .3f} s <br/>
-    RTF: {end - start: .3f}/{duration: .3f} = {(end - start)/duration:.3f} <br/>
+    RTF: {end - start: .3f}/{duration: .3f} = {rtf:.3f} <br/>
     """
+    if rtf > 1:
+        info += (
+            f"<br/>We are loading the model for the first run. "
+            "Please run again to measure the real RTF.<br/>"
+        )
+
     logging.info(info)
     logging.info(f"hyp:\n{hyp}")
 
@@ -171,12 +178,17 @@ title = "# Automatic Speech Recognition with Next-gen Kaldi"
 description = """
 This space shows how to do automatic speech recognition with Next-gen Kaldi.
 
+It is running on CPU within a docker container provided by Hugging Face.
+
 See more information by visiting the following links:
 
 - <https://github.com/k2-fsa/icefall>
 - <https://github.com/k2-fsa/sherpa>
 - <https://github.com/k2-fsa/k2>
 - <https://github.com/lhotse-speech/lhotse>
+
+If you want to deploy it locally, please see
+<https://k2-fsa.github.io/sherpa/>
 """
 
 # css style is copied from
@@ -199,6 +211,7 @@ def update_model_dropdown(language: str):
 
 demo = gr.Blocks(css=css)
 
+
 with demo:
     gr.Markdown(title)
     language_choices = list(language_to_models.keys())
@@ -245,6 +258,19 @@ with demo:
             uploaded_output = gr.Textbox(label="Recognized speech from uploaded file")
             uploaded_html_info = gr.HTML(label="Info")
 
+            gr.Examples(
+                examples=examples,
+                inputs=[
+                    language_radio,
+                    model_dropdown,
+                    decoding_method_radio,
+                    num_active_paths_slider,
+                    uploaded_file,
+                ],
+                outputs=[uploaded_output, uploaded_html_info],
+                fn=process_uploaded_file,
+            )
+
         with gr.TabItem("Record from microphone"):
             microphone = gr.Audio(
                 source="microphone",  # Choose between "microphone", "upload"
@@ -257,25 +283,39 @@ with demo:
             recorded_output = gr.Textbox(label="Recognized speech from recordings")
             recorded_html_info = gr.HTML(label="Info")
 
+            gr.Examples(
+                examples=examples,
+                inputs=[
+                    language_radio,
+                    model_dropdown,
+                    decoding_method_radio,
+                    num_active_paths_slider,
+                    microphone,
+                ],
+                outputs=[recorded_output, recorded_html_info],
+                fn=process_microphone,
+            )
+
         upload_button.click(
             process_uploaded_file,
             inputs=[
-                uploaded_file,
                 language_radio,
                 model_dropdown,
                 decoding_method_radio,
                 num_active_paths_slider,
+                uploaded_file,
             ],
             outputs=[uploaded_output, uploaded_html_info],
         )
+
         record_button.click(
             process_microphone,
             inputs=[
-                microphone,
                 language_radio,
                 model_dropdown,
                 decoding_method_radio,
                 num_active_paths_slider,
+                microphone,
             ],
             outputs=[recorded_output, recorded_html_info],
         )

examples.py

@@ -0,0 +1,156 @@
+#!/usr/bin/env python3
+#
+# Copyright      2022  Xiaomi Corp.        (authors: Fangjun Kuang)
+#
+# See LICENSE for clarification regarding multiple authors
+#
+# 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.
+examples = [
+    # librispeech
+    # https://huggingface.co/csukuangfj/icefall-asr-librispeech-pruned-transducer-stateless5-2022-05-13/tree/main/test_wavs
+    [
+        "English",
+        "csukuangfj/icefall-asr-librispeech-pruned-transducer-stateless3-2022-05-13",
+        "greedy_search",
+        4,
+        "./test_wavs/librispeech/1089-134686-0001.wav",
+    ],
+    [
+        "English",
+        "csukuangfj/icefall-asr-librispeech-pruned-transducer-stateless3-2022-05-13",
+        "greedy_search",
+        4,
+        "./test_wavs/librispeech/1221-135766-0001.wav",
+    ],
+    [
+        "English",
+        "csukuangfj/icefall-asr-librispeech-pruned-transducer-stateless3-2022-05-13",
+        "greedy_search",
+        4,
+        "./test_wavs/librispeech/1221-135766-0002.wav",
+    ],
+    # gigaspeech
+    [
+        "English",
+        "wgb14/icefall-asr-gigaspeech-pruned-transducer-stateless2",
+        "greedy_search",
+        4,
+        "./test_wavs/gigaspeech/1-minute-audiobook.opus",
+    ],
+    [
+        "English",
+        "wgb14/icefall-asr-gigaspeech-pruned-transducer-stateless2",
+        "greedy_search",
+        4,
+        "./test_wavs/gigaspeech/100-seconds-podcast.opus",
+    ],
+    [
+        "English",
+        "wgb14/icefall-asr-gigaspeech-pruned-transducer-stateless2",
+        "greedy_search",
+        4,
+        "./test_wavs/gigaspeech/100-seconds-youtube.opus",
+    ],
+    # wenetspeech
+    # https://huggingface.co/luomingshuang/icefall_asr_wenetspeech_pruned_transducer_stateless2/tree/main/test_wavs
+    [
+        "Chinese",
+        "luomingshuang/icefall_asr_wenetspeech_pruned_transducer_stateless2",
+        "greedy_search",
+        4,
+        "./test_wavs/wenetspeech/DEV_T0000000000.opus",
+    ],
+    [
+        "Chinese",
+        "luomingshuang/icefall_asr_wenetspeech_pruned_transducer_stateless2",
+        "greedy_search",
+        4,
+        "./test_wavs/wenetspeech/DEV_T0000000001.opus",
+    ],
+    [
+        "Chinese",
+        "luomingshuang/icefall_asr_wenetspeech_pruned_transducer_stateless2",
+        "greedy_search",
+        4,
+        "./test_wavs/wenetspeech/DEV_T0000000002.opus",
+    ],
+    # aishell2-A
+    # https://huggingface.co/yuekai/icefall-asr-aishell2-pruned-transducer-stateless5-A-2022-07-12/tree/main/test_wavs
+    [
+        "Chinese",
+        "yuekai/icefall-asr-aishell2-pruned-transducer-stateless5-A-2022-07-12",
+        "greedy_search",
+        4,
+        "./test_wavs/aishell2/ID0012W0030.wav",
+    ],
+    [
+        "Chinese",
+        "yuekai/icefall-asr-aishell2-pruned-transducer-stateless5-A-2022-07-12",
+        "greedy_search",
+        4,
+        "./test_wavs/aishell2/ID0012W0162.wav",
+    ],
+    [
+        "Chinese",
+        "yuekai/icefall-asr-aishell2-pruned-transducer-stateless5-A-2022-07-12",
+        "greedy_search",
+        4,
+        "./test_wavs/aishell2/ID0012W0215.wav",
+    ],
+    # aishell2-B
+    # https://huggingface.co/yuekai/icefall-asr-aishell2-pruned-transducer-stateless5-A-2022-07-12/tree/main/test_wavs
+    [
+        "Chinese",
+        "yuekai/icefall-asr-aishell2-pruned-transducer-stateless5-B-2022-07-12",
+        "greedy_search",
+        4,
+        "./test_wavs/aishell2/ID0012W0030.wav",
+    ],
+    [
+        "Chinese",
+        "yuekai/icefall-asr-aishell2-pruned-transducer-stateless5-B-2022-07-12",
+        "greedy_search",
+        4,
+        "./test_wavs/aishell2/ID0012W0162.wav",
+    ],
+    [
+        "Chinese",
+        "yuekai/icefall-asr-aishell2-pruned-transducer-stateless5-B-2022-07-12",
+        "greedy_search",
+        4,
+        "./test_wavs/aishell2/ID0012W0215.wav",
+    ],
+    # tal_csasr
+    # https://huggingface.co/luomingshuang/icefall_asr_tal-csasr_pruned_transducer_stateless5/tree/main/test_wavs
+    [
+        "Chinese+English",
+        "luomingshuang/icefall_asr_tal-csasr_pruned_transducer_stateless5",
+        "greedy_search",
+        4,
+        "./test_wavs/tal_csasr/210_36476_210_8341_1_1533271973_7057520_132.wav",
+    ],
+    [
+        "Chinese+English",
+        "luomingshuang/icefall_asr_tal-csasr_pruned_transducer_stateless5",
+        "greedy_search",
+        4,
+        "./test_wavs/tal_csasr/210_36476_210_8341_1_1533271973_7057520_138.wav",
+    ],
+    [
+        "Chinese+English",
+        "luomingshuang/icefall_asr_tal-csasr_pruned_transducer_stateless5",
+        "greedy_search",
+        4,
+        "./test_wavs/tal_csasr/210_36476_210_8341_1_1533271973_7057520_145.wav",
+    ],
+]

model.py

@@ -81,7 +81,7 @@ def _get_aishell2_pretrained_model(repo_id: str) -> OfflineAsr:
         "yuekai/icefall-asr-aishell2-pruned-transducer-stateless5-A-2022-07-12",  # noqa
         # context-size 2
         "yuekai/icefall-asr-aishell2-pruned-transducer-stateless5-B-2022-07-12",  # noqa
-    ]
+    ], repo_id
 
     nn_model_filename = _get_nn_model_filename(
         repo_id=repo_id,
@@ -102,12 +102,11 @@ def _get_aishell2_pretrained_model(repo_id: str) -> OfflineAsr:
 def _get_gigaspeech_pre_trained_model(repo_id: str) -> OfflineAsr:
     assert repo_id in [
         "wgb14/icefall-asr-gigaspeech-pruned-transducer-stateless2",
-    ]
+    ], repo_id
 
     nn_model_filename = _get_nn_model_filename(
-        # It is converted from https://huggingface.co/wgb14/icefall-asr-gigaspeech-pruned-transducer-stateless2  # noqa
-        repo_id="csukuangfj/icefall-asr-gigaspeech-pruned-transducer-stateless2",  # noqa
-        filename="cpu_jit-epoch-29-avg-11-torch-1.10.0.pt",
+        repo_id=repo_id,
+        filename="cpu_jit-iter-3488000-avg-20.pt",
     )
     bpe_model_filename = _get_bpe_model_filename(repo_id=repo_id)
 
@@ -124,7 +123,7 @@ def _get_gigaspeech_pre_trained_model(repo_id: str) -> OfflineAsr:
 def _get_librispeech_pre_trained_model(repo_id: str) -> OfflineAsr:
     assert repo_id in [
         "csukuangfj/icefall-asr-librispeech-pruned-transducer-stateless3-2022-05-13",  # noqa
-    ]
+    ], repo_id
 
     nn_model_filename = _get_nn_model_filename(
         repo_id=repo_id,
@@ -145,7 +144,7 @@ def _get_librispeech_pre_trained_model(repo_id: str) -> OfflineAsr:
 def _get_wenetspeech_pre_trained_model(repo_id: str):
     assert repo_id in [
         "luomingshuang/icefall_asr_wenetspeech_pruned_transducer_stateless2",
-    ]
+    ], repo_id
 
     nn_model_filename = _get_nn_model_filename(
         repo_id=repo_id,
@@ -166,7 +165,7 @@ def _get_wenetspeech_pre_trained_model(repo_id: str):
 def _get_tal_csasr_pre_trained_model(repo_id: str):
     assert repo_id in [
         "luomingshuang/icefall_asr_tal-csasr_pruned_transducer_stateless5",
-    ]
+    ], repo_id
 
     nn_model_filename = _get_nn_model_filename(
         repo_id=repo_id,
@@ -187,7 +186,7 @@ def _get_tal_csasr_pre_trained_model(repo_id: str):
 def _get_alimeeting_pre_trained_model(repo_id: str):
     assert repo_id in [
         "luomingshuang/icefall_asr_alimeeting_pruned_transducer_stateless2",
-    ]
+    ], repo_id
 
     nn_model_filename = _get_nn_model_filename(
         repo_id=repo_id,

test_wavs/aishell2/README.md

@@ -0,0 +1,2 @@
+Files are downloaded from
+https://huggingface.co/yuekai/icefall-asr-aishell2-pruned-transducer-stateless5-B-2022-07-12/tree/main/test_wavs

test_wavs/aishell2/trans.txt

@@ -0,0 +1,3 @@
+ID0012W0162 立法机关采纳了第二种意见
+ID0012W0215 大家都愿意牺牲自己的生命
+ID0012W0030 完全是典型的军事侵略

test_wavs/librispeech/README.md

@@ -0,0 +1,2 @@
+Files are downloaded from
+https://huggingface.co/csukuangfj/icefall-asr-librispeech-pruned-transducer-stateless5-2022-05-13/tree/main/test_wavs

test_wavs/librispeech/trans.txt

@@ -0,0 +1,3 @@
+1089-134686-0001 AFTER EARLY NIGHTFALL THE YELLOW LAMPS WOULD LIGHT UP HERE AND THERE THE SQUALID QUARTER OF THE BROTHELS
+1221-135766-0001 GOD AS A DIRECT CONSEQUENCE OF THE SIN WHICH MAN THUS PUNISHED HAD GIVEN HER A LOVELY CHILD WHOSE PLACE WAS ON THAT SAME DISHONOURED BOSOM TO CONNECT HER PARENT FOR EVER WITH THE RACE AND DESCENT OF MORTALS AND TO BE FINALLY A BLESSED SOUL IN HEAVEN
+1221-135766-0002 YET THESE THOUGHTS AFFECTED HESTER PRYNNE LESS WITH HOPE THAN APPREHENSION

test_wavs/tal_csasr/README.md

@@ -0,0 +1,2 @@
+Files are downloaded from
+https://huggingface.co/luomingshuang/icefall_asr_tal-csasr_pruned_transducer_stateless5/tree/main/test_wavs

test_wavs/wenetspeech/README.md

@@ -0,0 +1,2 @@
+Files are downloaded from
+https://huggingface.co/luomingshuang/icefall_asr_wenetspeech_pruned_transducer_stateless2/tree/main/test_wavs