setup jamo

游雁

2023-11-24 b5d3df75cf6462aa3bf42fd3c86fa2aa7f1c8a15

 funasr/datasets/dataset_jsonl.py

@@ -1,12 +1,48 @@
import torch
import json
import torch.distributed as dist
import numpy as np
import kaldiio
import librosa
import torchaudio
import time

class AudioDatasetJsonl(torch.utils.data.Dataset):
def load_audio(audio_path: str, fs: int=16000):
   audio = None
   if audio_path.startswith("oss:"):
      pass
   elif audio_path.startswith("odps:"):
      pass
   else:
      if ".ark:" in audio_path:
         audio = kaldiio.load_mat(audio_path)
      else:
         # audio, fs = librosa.load(audio_path, sr=fs)
         audio, fs = torchaudio.load(audio_path)
         audio = audio[0, :]
   return audio

def extract_features(data, date_type: str="sound", frontend=None):
   if date_type == "sound":

      if isinstance(data, np.ndarray):
         data = torch.from_numpy(data).to(torch.float32)
      data_len = torch.tensor([data.shape[0]]).to(torch.int32)
      feat, feats_lens = frontend(data[None, :], data_len)

      feat = feat[0, :, :]
   else:
      feat, feats_lens = torch.from_numpy(data).to(torch.float32), torch.tensor([data.shape[0]]).to(torch.int32)
   return feat, feats_lens
   
   def __init__(self, path, data_parallel_rank=0, data_parallel_size=1):
	

class IndexedDatasetJsonl(torch.utils.data.Dataset):
	
   def __init__(self, path):
      super().__init__()
      data_parallel_size = dist.get_world_size()
      # data_parallel_size = dist.get_world_size()
      data_parallel_size = 1
      contents = []
      with open(path, encoding='utf-8') as fin:
         for line in fin:
@@ -31,7 +67,8 @@
      self.contents = []
      total_num = len(contents)
      num_per_rank = total_num // data_parallel_size
      rank = dist.get_rank()
      # rank = dist.get_rank()
      rank = 0
      # import ipdb; ipdb.set_trace()
      self.contents = contents[rank * num_per_rank:(rank + 1) * num_per_rank]

@@ -41,3 +78,65 @@
   
   def __getitem__(self, index):
      return self.contents[index]


class AudioDataset(torch.utils.data.Dataset):
   def __init__(self, path, frontend=None, tokenizer=None, token_id_converter=None):

      super().__init__()
      self.indexed_dataset = IndexedDatasetJsonl(path)
      self.frontend = frontend.forward
      self.fs = 16000 if frontend is None else frontend.fs
      self.data_type = "sound"
      self.tokenizer = tokenizer
      self.token_id_converter = token_id_converter

      self.int_pad_value = -1
      self.float_pad_value = 0.0

	

	
   def __len__(self):
      return len(self.indexed_dataset)
	
   def __getitem__(self, index):
      item = self.indexed_dataset[index]
      # return item

      source = item["source"]
      data_src = load_audio(source, fs=self.fs)
      speech, speech_lengths = extract_features(data_src, self.data_type, self.frontend)
      target = item["target"]
      text = self.tokenizer.text2tokens(target)
      ids = self.token_id_converter.tokens2ids(text)
      ids_lengths = len(ids)
      text, text_lengths = torch.tensor(ids, dtype=torch.int64), torch.tensor([ids_lengths], dtype=torch.int32)

      return {"speech": speech,
              "speech_lengths": speech_lengths,
              "text": text,
              "text_lengths": text_lengths,
              }
	
	
   def collator(self, samples: list=None):
		
      # return samples
		
      outputs = {}
      for sample in samples:
         for key in sample.keys():
            if key not in outputs:
               outputs[key] = []
            outputs[key].append(sample[key])

      for key, data_list in outputs.items():
         if data_list[0].dtype == torch.int64:

            pad_value = self.int_pad_value
         else:
            pad_value = self.float_pad_value
         outputs[key] = torch.nn.utils.rnn.pad_sequence(data_list, batch_first=True, padding_value=pad_value)
      return outputs