Lukas Schmit

another bump!

would it be possible if the input is an mp3 to have `torchaudio.info` internally call `torchaudio.load` the file to get its metadata?

can also confirm that this just started happening in docker (`pytorchlightning/pytorch_lightning:base-cuda-py3.10-torch2.0-cuda11.7.1`) as of this morning. probably some latent dependency got bumped but haven't found which yet

adding that I'm fairly confident that [cython](https://pypi.org/project/Cython/) being bumped to `3.0.0` is causing this issue. timeline adds up & its a cython error. unfortunately pinning `cython

Also having issues, even for video/audio pairs from the same speaker in mead! it works perfectly for the same audio/video sample though. Is the `Inference_hyperlips.py` the correct thing to be...

Hi, I've also encountered this issue but with non-zero `num_workers`. Even more bizarre is I get this 75% of the way through a validation epoch when training with the latest...

@atong01 I think it should, but with the caveat that you might have to integrate the clean target through vector field (network) up to the current noisy timestep. Just using...

Hi, wanted to bump this! The readme indicates that fairseq2 binaries are available for torch 2.5.1. The exact command in the readme: ```sh pip install fairseq2 --extra-index-url https://fair.pkg.atmeta.com/fairseq2/whl/pt2.5.1/cu124 ``` will...