Mirco Ravanelli

I just fixed a backward incompatibility issue that was pretty problematic and caused the tests to fail. The bias of the DNN_block (used in all CRDNN) was set to false....

Thank you for @fpaissan for this PR! This contribution is crucial as we plan to release many more streamable models for various tasks like keyword spotting, speech recognition, and possibly...

I did some fixes and now training is running on my side. My next step is to check the streamable inference part.

https://huggingface.co/speechbrain/stream-vad-crdnn-libriparty/blob/main/README.md => It looks like the inference.py script only works with Apple devices, due to a limitation in torchaudio (in fact, it doesn't work on my Linux machine). If so,...

Thank you @mrouvier for pointing out this issue and sketching a solution. The problem with the proposed approach is that the field "wav" is not mandatory. There might be recipes...

One note of caution: this might lead to backward incompatibility changes as there is a mismatch between the old tokenizer and the new one. What do you think @lucadellalib and...

According to what I see, it seems there are some critical aspect. Firstly, when users click on the tutorial while navigating in the repository, Google Colab doesn't start, preventing users...

I think it is very important too. We can support at least LoRA and adapters. I'm not sure what could be the best way to go to support it elegantly....

@Adel-Moumen, what do you think about that?

Thank you @flexthink for your contribution! Having a model for MOS estimation is valuable for SpeechBrain. Here are some comments following an initial code inspection: 1. The README.md file is...