While the [docs](https://pytorch.org/audio/transforms.html#torchaudio.transforms.DownmixMono) state the default value for `channels_dim` is `True`, it seems to be set to `None` instead as shown [here](https://pytorch.org/audio/_modules/torchaudio/transforms.html#DownmixMono). This would yield to the default channel dimension as dim1: ```python channels_first = None ch_dim = int(not channels_first) print(ch_dim) > 1 ``` Reported in the forum [here](https://discuss.pytorch.org/t/how-does-torchaudio-transforms-downmixmono-work/41135). If you want, I could create a fast fix using `channels_first=True` as the default value or update the docs instead. Best, ptrblck