Julien Hauret

I also need this feature for ["Cnam-LMSSC/vibravox "](https://huggingface.co/datasets/Cnam-LMSSC/vibravox) EDIT: Upgrading to `2.19.0` fixed my problem thanks to [this PR](https://github.com/huggingface/datasets/pull/6754)

If I can answer your question, it comes from Shannon's source coding theorem: given an alphabet of symbols, you can choose shorter codes for more likely symbols to reduce the...

Hi and thanks for your reply, I am quite surprised to hear this! torchaudio is a great library that is very useful for the community! Is this only temporary?

Still on the [basic GAN tutorial](https://github.com/Lightning-AI/tutorials/blob/main/lightning_examples/basic-gan/gan.py), I spotted a few more track of improvements: - [on_validation_epoch_end](https://github.com/Lightning-AI/tutorials/blob/37e07988f981913721a613ac3ab40f53dcf6c9b2/lightning_examples/basic-gan/gan.py#L244) is ignored because `validation_step` is not defined. - After training, the quality of the...

Hi @Borda, I hope you're feeling better now. Yes, I'll do it the first week of August!

This point has also triggered my attention. The change has occured between [SEANet](https://arxiv.org/abs/2009.02095) that introduced $\mathbb{E}[\textrm{max}(0,1-D(\hat{x}))]$ and [MelGAN](https://arxiv.org/abs/1910.06711) that used $-\mathbb{E}[D(\hat{x})]$. After that, many similar papers have used it, including...

Thanks for pointing out the use of LSGAN loss in BigVGAN. Sorry if I was unclear. In fact, the values of the discriminators can be outside [-1, 1], but if...

I've fine-tuned Encodec's encoder and got interesting results for speech enhancement. I simply fed my low quality signal to a fine-tuned Encodec's encoder while feeding a frozen Encodec's encoder with...

I have "a" training code that I do not want to share publicly for the moment (written with Pytorch Lightning, which is very handy to get started quickly).