export

游雁

2023-03-29 cf00b4a13f5fdedda19c3cae214943fc28df52ac

 funasr/runtime/python/onnxruntime/funasr_onnx/vad_bin.py

@@ -23,7 +23,7 @@
                device_id: Union[str, int] = "-1",
                quantize: bool = False,
                intra_op_num_threads: int = 4,
                max_end_sil: int = 800,
                max_end_sil: int = None,
                ):
      
      if not Path(model_dir).exists():
@@ -43,14 +43,17 @@
      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
      self.max_end_sil = max_end_sil if max_end_sil is not None else config["vad_post_conf"]["max_end_silence_time"]
      self.encoder_conf = config["encoder_conf"]
   
   def prepare_cache(self, in_cache: list = []):
      if len(in_cache) > 0:
         return in_cache
		
      for i in range(4):
         cache = np.random.rand(1, 128, 19, 1).astype(np.float32)
      fsmn_layers = self.encoder_conf["fsmn_layers"]
      proj_dim = self.encoder_conf["proj_dim"]
      lorder = self.encoder_conf["lorder"]
      for i in range(fsmn_layers):
         cache = np.random.rand(1, proj_dim, lorder-1, 1).astype(np.float32)
         in_cache.append(cache)
      return in_cache