[doc]add gpu support for CLI

.gitignore

@@ -2,6 +2,7 @@
 __pycache__/
 *.py[cod]
 *$py.class
+*.egg-info
 
 # Visual Studio Code files
 .vscode
@@ -24,7 +25,7 @@ venv
 *.swo
 *.swp
 *.swm
-*~
+*~S
 
 # IPython notebook checkpoints
 .ipynb_checkpoints

README.md

@@ -25,6 +25,10 @@ pip install git+https://github.com/wenet-e2e/wespeaker.git
 $ wespeaker --task embedding --audio_file audio.wav --output_file embedding.txt
 $ wespeaker --task similarity --audio_file audio.wav --audio_file2 audio2.wav
 $ wespeaker --task diarization --audio_file audio.wav  # TODO
+
+# Add -g or --gpu to specify the gpu id to use, number < 0 means using CPU
+$ wespeaker --task embedding --audio_file audio.wav --output_file embedding.txt -g 0
+$ wespeaker --task similarity --audio_file audio.wav --audio_file2 audio2.wav --g 0
 ```
 
 **Python programming usage**:
@@ -33,6 +37,8 @@ $ wespeaker --task diarization --audio_file audio.wav  # TODO
 import wespeaker
 
 model = wespeaker.load_model('chinese')
+# set_gpu to enable the cuda inference, number < 0 means using CPU
+model.set_gpu(0)
 embedding = model.extract_embedding('audio.wav')
 similarity = model.compute_similarity('audio1.wav', 'audio2.wav')
 diar_result = model.diarize('audio.wav')  # TODO

wespeaker/cli/speaker.py

@@ -1,4 +1,5 @@
 # Copyright (c) 2023 Binbin Zhang ([email protected])
+#                    Shuai Wang ([email protected])
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -40,14 +41,23 @@ def __init__(self, model_dir: str):
         self.vad_model = vad.OnnxWrapper()
         self.table = {}
         self.resample_rate = 16000
-        self.apply_vad = True
+        self.apply_vad = False
+        self.device = torch.device('cpu')
 
     def set_resample_rate(self, resample_rate: int):
         self.resample_rate = resample_rate
 
     def set_vad(self, apply_vad: bool):
         self.apply_vad = apply_vad
 
+    def set_gpu(self, device_id: int):
+        if device_id >= 0:
+            device = 'cuda:{}'.format(device_id)
+        else:
+            device = 'cpu'
+        self.device = torch.device(device)
+        self.model = self.model.to(self.device)
+
     def extract_embedding(self, audio_path: str):
         pcm, sample_rate = torchaudio.load(audio_path, normalize=False)
         if self.apply_vad:
@@ -71,10 +81,11 @@ def extract_embedding(self, audio_path: str):
                             sample_frequency=16000)
         feats = feats - torch.mean(feats, 0)  # CMN
         feats = feats.unsqueeze(0)
+        feats = feats.to(self.device)
         self.model.eval()
         with torch.no_grad():
             _, outputs = self.model(feats)
-        embedding = outputs[0]
+        embedding = outputs[0].to(torch.device('cpu'))
         return embedding
 
     def compute_similarity(self, audio_path1: str, audio_path2: str) -> float:
@@ -150,6 +161,11 @@ def get_args():
     parser.add_argument('--vad',
                         action='store_true',
                         help='whether to do VAD or not')
+    parser.add_argument('-g',
+                        '--gpu',
+                        type=int,
+                        default=-1,
+                        help='which gpu to use (number <0 means using cpu)')
     parser.add_argument('--output_file',
                         help='output file to save speaker embedding')
     args = parser.parse_args()
@@ -161,6 +177,7 @@ def main():
     model = load_model(args.language)
     model.set_resample_rate(args.resample_rate)
     model.set_vad(args.vad)
+    model.set_gpu(args.gpu)
     if args.task == 'embedding':
         embedding = model.extract_embedding(args.audio_file)
         if embedding is not None: