update

Dockerfile

@@ -11,6 +11,7 @@ RUN pip install --upgrade pip
 
 RUN pip install --no-cache-dir --upgrade -r /code/requirements.txt -i https://pypi.tuna.tsinghua.edu.cn/simple
 
+# libk2_torch_api.so
 RUN export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/lib/python3.8/site-packages/k2/lib/
 
 # Set up a new user named "user" with user ID 1000

examples/wenet/infer.py

@@ -8,9 +8,14 @@ import sys
 pwd = os.path.abspath(os.path.dirname(__file__))
 sys.path.append(os.path.join(pwd, "../../"))
 
+import librosa
+import numpy as np
 import sherpa
+from scipy.io import wavfile
+import torch
+import torchaudio
 
-from project_settings import project_path
+from project_settings import project_path, temp_directory
 
 
 def get_args():
@@ -20,6 +25,12 @@ def get_args():
         default=(project_path / "pretrained_models/huggingface/csukuangfj/wenet-chinese-model").as_posix(),
         type=str
     )
+    parser.add_argument(
+        "--filename",
+        default=(project_path / "data/test_wavs/paraformer-zh/四川话.wav").as_posix(),
+        type=str
+    )
+    parser.add_argument("--sample_rate", default=16000, type=int)
     args = parser.parse_args()
     return args
 
@@ -32,7 +43,7 @@ def main():
     tokens_filename = model_dir / "units.txt"
 
     feat_config = sherpa.FeatureConfig(normalize_samples=False)
-    feat_config.fbank_opts.frame_opts.samp_freq = 16000
+    feat_config.fbank_opts.frame_opts.samp_freq = args.sample_rate
     feat_config.fbank_opts.mel_opts.num_bins = 80
     feat_config.fbank_opts.frame_opts.dither = 0
 
@@ -46,7 +57,28 @@ def main():
     )
 
     recognizer = sherpa.OfflineRecognizer(config)
-    print(recognizer)
+
+    signal, sample_rate = librosa.load(args.filename, sr=args.sample_rate)
+    signal *= 32768.0
+    signal = np.array(signal, dtype=np.int16)
+
+    temp_file = temp_directory / "temp.wav"
+    wavfile.write(
+        temp_file.as_posix(),
+        rate=args.sample_rate,
+        data=signal
+    )
+
+    s = recognizer.create_stream()
+
+    s.accept_wave_file(
+        temp_file.as_posix()
+    )
+    recognizer.decode_stream(s)
+
+    text = s.result.text.strip()
+    text = text.lower()
+    print("text: {}".format(text))
     return
 
 

project_settings.py

@@ -7,6 +7,9 @@ from pathlib import Path
 project_path = os.path.abspath(os.path.dirname(__file__))
 project_path = Path(project_path)
 
+temp_directory = project_path / "temp"
+temp_directory.mkdir(parents=True, exist_ok=True)
+
 
 if __name__ == '__main__':
     pass

requirements.txt

@@ -1,6 +1,7 @@
 gradio==4.29.0
 torch==1.13.1
 torchaudio==0.13.1
+librosa==0.8.1
 
 data/wheels/k2-1.23.4.dev20230130+cpu.torch1.13.1-cp38-cp38-linux_x86_64.whl
 data/wheels/k2_sherpa-1.1-cp38-cp38-linux_x86_64.whl