Unverified Commit ab4fc99a authored by djqualia's avatar djqualia Committed by GitHub
Browse files

Convert stereo/multi-channel audio to mono 

It is annoying when training on large data sets for one non-mono file to cause an exception (einops.EinopsError: Shape mismatch, 2 != 1) and the whole training pipeline to crash.  Weeding out such files can take some time.  This will avoid such crashes and handle them gracefully...
parent ba7dcd68

audiolm_pytorch/data.py

+4 −0
Original line number Diff line number Diff line
@@ -68,6 +68,10 @@ class SoundDataset(Dataset): 


        assert data.numel() > 0, f'one of your audio file ({file}) is empty. please remove it from your folder'



        if data.shape[0] > 1:

            # the audio has more than 1 channel, convert to mono

            data = torch.mean(data, dim=0).unsqueeze(0)



        num_outputs = len(self.target_sample_hz)

        data = cast_tuple(data, num_outputs)