RuntimeError: "unfolded2d_copy" not implemented for 'Half'

[GauravCodesAI]

I'm Trying To Transcribe an Audio File.

Code:

model = whisper.load_model("small")

def whisper_scribe(audio):
    audio_path = os.path.join(test_dir, audio)

    # load audio and pad/trim it to fit 30 seconds
    audio, sr = librosa.load(audio_path, sr=16000, offset=15)
#     audio = whisper.load_audio(audio_path) # Specified File Not Found Error

    audio = whisper.pad_or_trim(audio)

    # make log-Mel spectrogram
    mel = whisper.log_mel_spectrogram(audio)

    # detect the spoken language
    _, probs = model.detect_language(mel)

    print(f"Detected language: {max(probs, key=probs.get)}")

    # decode the audio
    options = whisper.DecodingOptions()
    result = whisper.decode(model, mel, options)

    # print the recognized text
    print(result)

Error:

Detected language: en
---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
<ipython-input-4-784a5feb6267> in <module>
----> 1 whisper_scribe('test31.wav')

<ipython-input-3-6c02e761d53a> in whisper_scribe(audio)
     20     # decode the audio
     21     options = whisper.DecodingOptions()
---> 22     result = whisper.decode(model, mel, options)
     23 
     24     # print the recognized text

C:\ProgramData\Anaconda3\lib\site-packages\torch\autograd\grad_mode.py in decorate_context(*args, **kwargs)
     26         def decorate_context(*args, **kwargs):
     27             with self.__class__():
---> 28                 return func(*args, **kwargs)
     29         return cast(F, decorate_context)
     30 

C:\ProgramData\Anaconda3\lib\site-packages\whisper\decoding.py in decode(model, mel, options)
    699         mel = mel.unsqueeze(0)
    700 
--> 701     result = DecodingTask(model, options).run(mel)
    702 
    703     if single:

C:\ProgramData\Anaconda3\lib\site-packages\torch\autograd\grad_mode.py in decorate_context(*args, **kwargs)
     26         def decorate_context(*args, **kwargs):
     27             with self.__class__():
---> 28                 return func(*args, **kwargs)
     29         return cast(F, decorate_context)
     30 

C:\ProgramData\Anaconda3\lib\site-packages\whisper\decoding.py in run(self, mel)
    615         n_audio: int = mel.shape[0]
    616 
--> 617         audio_features: Tensor = self._get_audio_features(mel)  # encoder forward pass
    618         tokens: Tensor = torch.tensor([self.initial_tokens]).repeat(n_audio, 1)
    619 

C:\ProgramData\Anaconda3\lib\site-packages\whisper\decoding.py in _get_audio_features(self, mel)
    559             audio_features = mel
    560         else:
--> 561             audio_features = self.model.encoder(mel)
    562 
    563         if audio_features.dtype != (torch.float16 if self.options.fp16 else torch.float32):

C:\ProgramData\Anaconda3\lib\site-packages\torch\nn\modules\module.py in _call_impl(self, *input, **kwargs)
   1049         if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
   1050                 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1051             return forward_call(*input, **kwargs)
   1052         # Do not call functions when jit is used
   1053         full_backward_hooks, non_full_backward_hooks = [], []

C:\ProgramData\Anaconda3\lib\site-packages\whisper\model.py in forward(self, x)
    146             the mel spectrogram of the audio
    147         """
--> 148         x = F.gelu(self.conv1(x))
    149         x = F.gelu(self.conv2(x))
    150         x = x.permute(0, 2, 1)

C:\ProgramData\Anaconda3\lib\site-packages\torch\nn\modules\module.py in _call_impl(self, *input, **kwargs)
   1049         if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
   1050                 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1051             return forward_call(*input, **kwargs)
   1052         # Do not call functions when jit is used
   1053         full_backward_hooks, non_full_backward_hooks = [], []

C:\ProgramData\Anaconda3\lib\site-packages\torch\nn\modules\conv.py in forward(self, input)
    296 
    297     def forward(self, input: Tensor) -> Tensor:
--> 298         return self._conv_forward(input, self.weight, self.bias)
    299 
    300 

C:\ProgramData\Anaconda3\lib\site-packages\whisper\model.py in _conv_forward(self, x, weight, bias)
     41 class Conv1d(nn.Conv1d):
     42     def _conv_forward(self, x: Tensor, weight: Tensor, bias: Optional[Tensor]) -> Tensor:
---> 43         return super()._conv_forward(
     44             x, weight.to(x.dtype), None if bias is None else bias.to(x.dtype)
     45         )

C:\ProgramData\Anaconda3\lib\site-packages\torch\nn\modules\conv.py in _conv_forward(self, input, weight, bias)
    292                             weight, bias, self.stride,
    293                             _single(0), self.dilation, self.groups)
--> 294         return F.conv1d(input, weight, bias, self.stride,
    295                         self.padding, self.dilation, self.groups)
    296 

RuntimeError: "unfolded2d_copy" not implemented for 'Half'

[zanjabil2502]

you can edit like this:
options = whisper.DecodingOptions(fp16=False)

[GauravCodesAI]

Thank You!