update

examples/add_punctuation/add_punctuation.py

@@ -0,0 +1,43 @@
+#!/usr/bin/python3
+# -*- coding: utf-8 -*-
+import argparse
+
+import sherpa_onnx
+
+from project_settings import project_path
+
+
+def get_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--model_file",
+        default=(project_path / "pretrained_models/huggingface/csukuangfj/sherpa-onnx-punct-ct-transformer-zh-en-vocab272727-2024-04-12/model.onnx").as_posix(),
+        type=str
+    )
+    parser.add_argument(
+        "--text",
+        default="i'm a google virtual assistant recording this call for the person you're trying to reach before i try to connect you can ask what you're calling about",
+        type=str
+    )
+    args = parser.parse_args()
+    return args
+
+
+def main():
+    args = get_args()
+
+    config = sherpa_onnx.OfflinePunctuationConfig(
+        model=sherpa_onnx.OfflinePunctuationModelConfig(
+            ct_transformer=args.model_file
+        ),
+    )
+
+    punctuation_model = sherpa_onnx.OfflinePunctuation(config)
+
+    text = punctuation_model.add_punctuation(args.text)
+    print("text: {}".format(text))
+    return
+
+
+if __name__ == '__main__':
+    main()

examples/add_punctuation/download_model.py

@@ -0,0 +1,58 @@
+#!/usr/bin/python3
+# -*- coding: utf-8 -*-
+import argparse
+import os
+from pathlib import Path
+import sys
+
+pwd = os.path.abspath(os.path.dirname(__file__))
+sys.path.append(os.path.join(pwd, "../../"))
+
+import huggingface_hub
+
+from project_settings import project_path
+
+
+def get_args():
+    parser = argparse.ArgumentParser()
+
+    parser.add_argument(
+        "--repo_id",
+        default="csukuangfj/sherpa-onnx-punct-ct-transformer-zh-en-vocab272727-2024-04-12",
+        type=str
+    )
+    parser.add_argument("--model_filename", default="model.onnx", type=str)
+    parser.add_argument("--model_sub_folder", default=".", type=str)
+
+    parser.add_argument(
+        "--pretrained_model_dir",
+        default=(project_path / "pretrained_models").as_posix(),
+        type=str
+    )
+    args = parser.parse_args()
+    return args
+
+
+def main():
+    args = get_args()
+
+    pretrained_model_dir = Path(args.pretrained_model_dir)
+    pretrained_model_dir.mkdir(exist_ok=True)
+
+    repo_id: Path = Path(args.repo_id)
+    local_model_dir = pretrained_model_dir / "huggingface" / repo_id
+    local_model_dir.mkdir(parents=True, exist_ok=True)
+
+    print("download model")
+    model_filename = huggingface_hub.hf_hub_download(
+        repo_id=args.repo_id,
+        filename=args.model_filename,
+        subfolder=args.model_sub_folder,
+        local_dir=local_model_dir.as_posix(),
+    )
+    print(model_filename)
+    return
+
+
+if __name__ == "__main__":
+    main()

examples/gradio_client/whisper_large_v3.py

@@ -0,0 +1,37 @@
+#!/usr/bin/python3
+# -*- coding: utf-8 -*-
+import argparse
+
+from gradio_client import Client, file
+
+from project_settings import project_path
+
+
+def get_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--filename",
+        default=(project_path / "data/test_wavs/paraformer-zh/si_chuan_hua.wav").as_posix(),
+        type=str
+    )
+    args = parser.parse_args()
+    return args
+
+
+def main():
+    args = get_args()
+
+    filename = args.filename
+
+    client = Client("hf-audio/whisper-large-v3")
+    result = client.predict(
+        inputs=file(filename),
+        task="transcribe",
+        api_name="/predict"
+    )
+    print(result)
+    return
+
+
+if __name__ == '__main__':
+    main()

main.py

@@ -148,10 +148,20 @@ def process(
                                        filename=out_filename.as_posix(),
                                        )
 
+    # load_punctuation_model
+    if add_punctuation == "Yes":
+        local_model_dir = pretrained_model_dir / "huggingface" / md5_encrypt("csukuangfj/sherpa-onnx-punct-ct-transformer-zh-en-vocab272727-2024-04-12")
+        punctuation_model = nn_models.load_punctuation_model(
+            local_model_dir=local_model_dir,
+            nn_model_file="model.onnx",
+            nn_model_file_sub_folder=".",
+        )
+        text = punctuation_model.add_punctuation(text)
+
+    # statistics
     date_time = now.strftime("%Y-%m-%d %H:%M:%S.%f")
     end = time.time()
 
-    # statistics
     metadata = torchaudio.info(out_filename.as_posix())
     duration = metadata.num_frames / 16000
     rtf = (end - start) / duration

toolbox/k2_sherpa/nn_models.py

@@ -764,7 +764,7 @@ def load_sherpa_onnx_online_recognizer_from_paraformer(encoder_model_file: str,
 def load_recognizer(local_model_dir: Path,
                     decoding_method: str = "greedy_search",
                     num_active_paths: int = 4,
-                    **kwargs
+                    **kwargs,
                     ):
     if not local_model_dir.exists():
         download_model(
@@ -839,5 +839,29 @@ def load_recognizer(local_model_dir: Path,
     return recognizer
 
 
+def load_punctuation_model(local_model_dir: Path,
+                           nn_model_file: str,
+                           nn_model_file_sub_folder: str,
+                           ):
+    if not local_model_dir.exists():
+        download_model(
+            local_model_dir=local_model_dir.as_posix(),
+            nn_model_file=nn_model_file,
+            nn_model_file_sub_folder=nn_model_file_sub_folder,
+        )
+
+    nn_model_file = (local_model_dir / nn_model_file_sub_folder / nn_model_file).as_posix()
+
+    config = sherpa_onnx.OfflinePunctuationConfig(
+        model=sherpa_onnx.OfflinePunctuationModelConfig(
+            ct_transformer=nn_model_file
+        ),
+    )
+
+    punctuation_model = sherpa_onnx.OfflinePunctuation(config)
+
+    return punctuation_model
+
+
 if __name__ == "__main__":
     pass