Fix Normalization

processing_moss_tts.py

@@ -687,7 +687,7 @@ class MossTTSDelayProcessor(ProcessorMixin):
                 audio_start_idx = int(audio_start_idx_t.item())
                 audio_end_idx = int(audio_end_idx_t.item())
                 delay_audio_codes = self.apply_delay_pattern(
-                    audio_ocodes, self.model_config.audio_pad_code
+                    audio_codes, self.model_config.audio_pad_code
                 )
                 pad_codes = torch.full(
                     (audio_start_idx - prefix_idx + 1, n_vq),
@@ -762,7 +762,7 @@ class MossTTSDelayProcessor(ProcessorMixin):
         if breaks.numel() == 0:
             segments_idx = [idx]
         else:
-            segments_idx = torch.split(idx, breaks.tolist())
+            segments_idx = torch.split(idx, breaks)
 
         audio_codes_list = [audio_codes[s] for s in segments_idx]