Socket通信中pipeline执行speaker-verification任务出现choose a window size 400 that is [2, 0]报错

[eugeneYz]

🐛 Bug

场景socket通信 ，API：pipelines，Task：'speaker-verification'的，方法：pipeline[wav1,wav2] ，报错：choose a window size 400 that is [2, 0] Debug不报错，RUN会报错。 方法拿出来单独测试就没问题。

To Reproduce

Steps to reproduce the behavior (always include the command you ran):

1. Run cmd '....'
2. See error

Code sample

        self.pipline = pipeline(
            task='speaker-verification',
            model=self.model_path,
        )
        wav_dict = {file_path: None for file_path in wav_files}
        for key, value in wav_dict.items():
            time.sleep(0.1)
            print("说话人确认计算中...")
            print("key:", key)
            print("string_path:", string_path)
            result = self.pipline([string_path, key])
            wav_dict[key] = result
            print("key:", key)
            print("res:", wav_dict[key])
            print("\n",)

Environment

• OS (e.g., Win10):
• FunASR Version (e.g., 1.0.14):
• ModelScope Version (e.g., 1.13.1):
• PyTorch Version (e.g., 2.1.2):
• How you installed funasr (pip, source):
• Python version:3.8

[eugeneYz]

不清楚该报错的含义和背后的原理，无从下手Debug，请帮忙，感谢！