paraformer streaming bugfix

游雁

2024-03-18 cbe2ea7e07cbf364827bd89cefc42b3f643ea3be

 runtime/python/onnxruntime/funasr_onnx/vad_bin.py

@@ -31,7 +31,8 @@
                quantize: bool = False,
                intra_op_num_threads: int = 4,
                max_end_sil: int = None,
                cache_dir: str = None
                cache_dir: str = None,
                **kwargs
                ):
      
      if not Path(model_dir).exists():
@@ -62,9 +63,9 @@
                  "\npip3 install -U funasr -i https://mirror.sjtu.edu.cn/pypi/web/simple"
         
         model = AutoModel(model=model_dir)
         model_dir = model.export(type="onnx", quantize=quantize)
      config_file = os.path.join(model_dir, 'vad.yaml')
      cmvn_file = os.path.join(model_dir, 'vad.mvn')
         model_dir = model.export(type="onnx", quantize=quantize, **kwargs)
      config_file = os.path.join(model_dir, 'config.yaml')
      cmvn_file = os.path.join(model_dir, 'am.mvn')
      config = read_yaml(config_file)
      
      self.frontend = WavFrontend(
@@ -73,8 +74,8 @@
      )
      self.ort_infer = OrtInferSession(model_file, device_id, intra_op_num_threads=intra_op_num_threads)
      self.batch_size = batch_size
      self.vad_scorer = E2EVadModel(config["vad_post_conf"])
      self.max_end_sil = max_end_sil if max_end_sil is not None else config["vad_post_conf"]["max_end_silence_time"]
      self.vad_scorer = E2EVadModel(config["model_conf"])
      self.max_end_sil = max_end_sil if max_end_sil is not None else config["model_conf"]["max_end_silence_time"]
      self.encoder_conf = config["encoder_conf"]
   
   def prepare_cache(self, in_cache: list = []):
@@ -196,7 +197,8 @@
                quantize: bool = False,
                intra_op_num_threads: int = 4,
                max_end_sil: int = None,
                cache_dir: str = None
                cache_dir: str = None,
                **kwargs
                ):
      if not Path(model_dir).exists():
         try:
@@ -226,10 +228,10 @@
                  "\npip3 install -U funasr -i https://mirror.sjtu.edu.cn/pypi/web/simple"
         
         model = AutoModel(model=model_dir)
         model_dir = model.export(type="onnx", quantize=quantize)
         model_dir = model.export(type="onnx", quantize=quantize, **kwargs)
         
      config_file = os.path.join(model_dir, 'vad.yaml')
      cmvn_file = os.path.join(model_dir, 'vad.mvn')
      config_file = os.path.join(model_dir, 'config.yaml')
      cmvn_file = os.path.join(model_dir, 'am.mvn')
      config = read_yaml(config_file)
      
      self.frontend = WavFrontendOnline(
@@ -238,8 +240,8 @@
      )
      self.ort_infer = OrtInferSession(model_file, device_id, intra_op_num_threads=intra_op_num_threads)
      self.batch_size = batch_size
      self.vad_scorer = E2EVadModel(config["vad_post_conf"])
      self.max_end_sil = max_end_sil if max_end_sil is not None else config["vad_post_conf"]["max_end_silence_time"]
      self.vad_scorer = E2EVadModel(config["model_conf"])
      self.max_end_sil = max_end_sil if max_end_sil is not None else config["model_conf"]["max_end_silence_time"]
      self.encoder_conf = config["encoder_conf"]
   
   def prepare_cache(self, in_cache: list = []):