update

ultravox_processing.py

@@ -171,6 +171,7 @@ class UltravoxProcessor(transformers.ProcessorMixin):
                 text, list
             ), "Text must be a list."
             processed_text = []
+            data["audio_token_start_idx"] = []
             for t in text:
                 if self.audio_placeholder in t:
                     if "audio_token_len" not in data:
@@ -184,7 +185,7 @@ class UltravoxProcessor(transformers.ProcessorMixin):
                             add_special_tokens=False,
                         )
                     )
-                    data["audio_token_start_idx"] = [start_idx]
+                    data["audio_token_start_idx"].append(start_idx)
 
                     # Replace the audio placeholder with the audio token.
                     #   e.g. "Transcribe\n<|audio|>" -> "Transcribe </s></s></s></s></s></s></s></s>"