funasr/datasets/audio_datasets/samplers.py
@@ -13,6 +13,7 @@ buffer_size: int = 30, drop_last: bool = False, shuffle: bool = True, is_training: bool = True, **kwargs): self.drop_last = drop_last @@ -24,10 +25,10 @@ self.buffer_size = buffer_size self.max_token_length = kwargs.get("max_token_length", 5000) self.shuffle_idx = np.arange(self.total_samples) self.shuffle = shuffle self.shuffle = shuffle and is_training def __len__(self): return self.total_samples return (self.total_samples-1) // self.batch_size + 1 def set_epoch(self, epoch): np.random.seed(epoch)
