RuntimeError: Sizes of tensors must match except in dimension 0. Expected size 80000 but got size 79659 for tensor number 30 in the list

[Pkoiralap]

I was experiencing this issue and was able to solve it using guard rails around the waveform creation process. However, I believe there should be a better way to do this and that's why I am not creating a PR with the fix. If you think this fix works, I can submit a PR.

So in the audio/pipelines/speaker_diarization.py file. i.e. this bit here:

# chunk: Segment(t, t + duration)
# masks: (num_frames, local_num_speakers) np.ndarray
waveform, _ = self._audio.crop(
    file,
    chunk,
    duration=duration,
    mode="pad",
)
# waveform: (1, num_samples) torch.Tensor
...

creates a tensor of a different size than what is expected. The simple fix I am using for now is:

# chunk: Segment(t, t + duration)
# masks: (num_frames, local_num_speakers) np.ndarray
waveform, _ = self._audio.crop(
    file,
    chunk,
    duration=duration,
    mode="pad",
)
# waveform: (1, num_samples) torch.Tensor
if waveform.shape[1] < num_samples:
    pad_num = int(num_samples - waveform.shape[1])
    waveform = torch.nn.functional.pad(waveform, (0, pad_num), "constant", 0)

This is hacky, but it works for me. Also because I didn't have a lot of time reading through the code, I am not sure if this is the right fix for the problem. As I said earlier, depending on whether this is a 'good' fix, I can create a PR accordingly.

Thanks

[github-actions]

Thank you for your issue. Give us a little time to review it.

PS. You might want to check the FAQ if you haven't done so already.

This is an automated reply, generated by FAQtory

[hbredin]

Would you mind narrowing down when this problem occurs?

This is most likely a bug in the implementation of the mode="pad" option which is already supposed to handle this corner case...

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.