T

qwen_tts/inference/qwen3_tts_model.py

@@ -621,27 +621,8 @@ class Qwen3TTSModel:
             **gen_kwargs,
         )
 
-        codes_for_decode = []
-        for i, codes in enumerate(talker_codes_list):
-            ref_code_list = voice_clone_prompt_dict.get("ref_code", None)
-            if ref_code_list is not None and ref_code_list[i] is not None:
-                codes_for_decode.append(torch.cat([ref_code_list[i].to(codes.device), codes], dim=0))
-            else:
-                codes_for_decode.append(codes)
-
         logger.info("模型推理完成，正在解码音频...")
-        wavs_all, fs = self.model.speech_tokenizer.decode([{"audio_codes": c} for c in codes_for_decode])
-
-        wavs_out: List[np.ndarray] = []
-        for i, wav in enumerate(wavs_all):
-            ref_code_list = voice_clone_prompt_dict.get("ref_code", None)
-            if ref_code_list is not None and ref_code_list[i] is not None:
-                ref_len = int(ref_code_list[i].shape[0])
-                total_len = int(codes_for_decode[i].shape[0])
-                cut = int(ref_len / max(total_len, 1) * wav.shape[0])
-                wavs_out.append(wav[cut:])
-            else:
-                wavs_out.append(wav)
+        wavs_out, fs = self.model.speech_tokenizer.decode([{"audio_codes": c} for c in talker_codes_list])
 
         return wavs_out, fs