train finetune

游雁

2024-02-20 45d9ccafef8d5feade0665d52ba5a32ea62b938d

 funasr/datasets/audio_datasets/preprocessor.py

@@ -26,8 +26,10 @@
         return waveform
      speed = random.choice(self.speed_perturb)
      if speed != 1.0:
         if not isinstance(waveform, torch.Tensor):
            waveform = torch.tensor(waveform)
         waveform, _ = torchaudio.sox_effects.apply_effects_tensor(
            torch.tensor(waveform).view(1, -1), fs, [['speed', str(speed)], ['rate', str(fs)]])
            waveform.view(1, -1), fs, [['speed', str(speed)], ['rate', str(fs)]])
         waveform = waveform.view(-1)
         
      return waveform
@@ -41,43 +43,9 @@
                **kwargs):
      super().__init__()
      
      self.seg_dict = None
      if seg_dict is not None:
         self.seg_dict = {}
         with open(seg_dict, "r", encoding="utf8") as f:
            lines = f.readlines()
         for line in lines:
            s = line.strip().split()
            key = s[0]
            value = s[1:]
            self.seg_dict[key] = " ".join(value)
      self.text_cleaner = TextCleaner(text_cleaner)
      self.split_with_space = split_with_space
   
   def forward(self, text, **kwargs):
      if self.seg_dict is not None:
         text = self.text_cleaner(text)
         if self.split_with_space:
            tokens = text.strip().split(" ")
            if self.seg_dict is not None:
               text = seg_tokenize(tokens, self.seg_dict)

      text = self.text_cleaner(text)
		
      return text

def seg_tokenize(txt, seg_dict):
   pattern = re.compile(r'^[\u4E00-\u9FA50-9]+$')
   out_txt = ""
   for word in txt:
      word = word.lower()
      if word in seg_dict:
         out_txt += seg_dict[word] + " "
      else:
         if pattern.match(word):
            for char in word:
               if char in seg_dict:
                  out_txt += seg_dict[char] + " "
               else:
                  out_txt += "<unk>" + " "
         else:
            out_txt += "<unk>" + " "
   return out_txt.strip().split()