fix encodec collator

run_stable_speech_training.py

@@ -495,13 +495,10 @@ class DataCollatorEncodecWithPadding:
     def __call__(self, features: List[Dict[str, Union[List[int], torch.Tensor]]]) -> Dict[str, torch.Tensor]:
         # split inputs and labels since they have to be of different lengths and need
         # different padding methods
-        audios = [torch.tensor(feature[self.audio_column_name]).squeeze() for feature in features]
+        audios = [feature[self.audio_column_name]["array"] for feature in features]
         len_audio = [len(audio) for audio in audios]
         
-        batch = self.feature_extractor(audios, return_tensors="pt", padding="longest", return_attention_mask=True)
-
-        batch[self.feature_extractor_input_name] = batch[self.feature_extractor_input_name].unsqueeze(1) # add mono-channel 
-        batch["padding_mask"] = batch.pop("attention_mask")   
+        batch = self.feature_extractor(audios, return_tensors="pt", padding="longest")
         batch["len_audio"] = torch.tensor(len_audio).unsqueeze(1)
         return batch
 
@@ -1083,17 +1080,15 @@ def main():
         # this is a trick to avoid to rewrite the entire audio column which takes ages
         cols_to_remove = [col for col  in next(iter(raw_datasets.values())).column_names if col != target_audio_column_name]
         for split in raw_datasets:
-            vectorized_datasets[split] = concatenate_datasets([raw_datasets[split].remove_columns(cols_to_remove), tmp_datasets[split]], axis=1)
+            raw_datasets[split] = concatenate_datasets([raw_datasets[split].remove_columns(cols_to_remove), tmp_datasets[split]], axis=1)
 
-        # TODO: remove
-        logger.info(f"Vectorized datasets {vectorized_datasets}")
 
         with accelerator.main_process_first():
             def is_audio_in_length_range(length):
                 return length > min_target_length and length < max_target_length
 
             # filter data that is shorter than min_target_length
-            vectorized_datasets = vectorized_datasets.filter(
+            vectorized_datasets = raw_datasets.filter(
                 is_audio_in_length_range,
                 num_proc=num_workers,
                 input_columns=["target_length"],