Python中Whisper库model.transcribe报错tuple.index(x): x not in tuple

问题背景 #

我开发了一款支持用户选择音频文件并指定语言进行转录的程序，但运行过程中触发如下错误。

错误日志 #

ERROR:root:Error at division
Traceback (most recent call last):
  File "C:\Users\...\AppData\Local\Temp\ipykernel_27364\1513539390.py", line 136, in transcribir
    result = model.transcribe(audio, language=f"{actual_language}", fp16=False, verbose=True)
  File "C:\Users\...\anaconda3\envs\python39\lib\site-packages\whisper\transcribe.py", line 181, in transcribe
    ):
  File "C:\Users\...\anaconda3\envs\python39\lib\site-packages\whisper\transcribe.py", line 117, in decode_with_fallback
  File "C:\Users\...\anaconda3\envs\python39\lib\site-packages\torch\utils\_contextlib.py", line 115, in decorate_context
    return func(*args, **kwargs)
  File "C:\Users\...\anaconda3\envs\python39\lib\site-packages\whisper\decoding.py", line 705, in decode
    if completed or tokens.shape[-1] > self.n_ctx:
  File "C:\Users\...\anaconda3\envs\python39\lib\site-packages\whisper\decoding.py", line 468, in __init__
    if last_was_timestamp:
ValueError: tuple.index(x): x not in tuple

当前功能代码 #

actual_language=combo.get()

archivo=filedialog.askopenfilename(initialdir = "./",
             title = "Seleccione archivo",filetypes = (("all files","*.*"),("mp3 files","*.mp3"),
             ("wav files","*.wav")))

torch.cuda.is_available()
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"

model = whisper.load_model("medium")

if archivo!="":
    try:
                
        audio = whisper.load_audio(archivo)
        audio_trim = whisper.pad_or_trim(audio)
        
        mel = whisper.log_mel_spectrogram(audio_trim).to(model.device)
        
        result = model.transcribe(audio, language=f"{actual_language}", fp16=False, verbose=True)
        
        print(result["text"])

    except Exception as e:
        actual_status.set("¡Error!")
        logging.error('Error at %s', 'division', exc_info=e)

已尝试的操作 #

• 添加语言选择功能前，使用自动语言识别仍触发相同错误，尝试代码：

  result = model.transcribe(audio)
  

  或

  audio = whisper.load_audio(file)
  audio_trim = whisper.pad_or_trim(audio)
  
  mel = whisper.log_mel_spectrogram(audio_trim).to(model.device)
  
  _, probs = model.detect_language(mel)
  
  options = whisper.DecodingOptions()
  result = whisper.decode(model, mel, options)
  

• 添加语言选择后，简化transcribe调用参数：

  result = model.transcribe(archivo,language=f"{actual_language}")
  

环境信息 #

• Python版本：3.9
• Whisper版本：已升级至最新版

内容的提问来源于stack exchange，提问作者Nacho Delgado