clone_model = CosyVoice3(os.path.join("checkpoints", "Fun-CosyVoice3-0.5B"), load_trt=True, load_vllm=True,
fp16=True)
for chunk in clone_model.inference_instruct2(text_content, instruct_text, local_path, "", False,
speed, True):
生成的音频如果保存为MP3会报错nan,如果保存为wav,全是静音。这是什么问题,官方的代码有bug吗?