Pavel Denisov

Here is also C++ implementation: https://github.com/beschulz/wav2json

Hi @wendywong343! You can see an example of usage of tine-tuned `facebook/mbart-large-50-many-to-many-mmt` in this PR: https://github.com/espnet/espnet/pull/4099 It's still very experimental and there might be issues, but it works for me...

Sorry if my comment was confusing, there is no SLU example here yet, just the general code changes. So maybe it's not ready for full review now, but any comments...

@sw005320 @siddhu001 It took forever to make it to work, but now I think it's ready for review. That's a lot of changes, please let me know what you think....

Hi @siddhu001 @sw005320, thanks a lot for your review and comments! I made the suggested changes. I have a bit of concern about reproducibility of the currently shown results, because...

Hi @siddhu001 @sw005320! > I have a bit of concern about reproducibility of the currently shown results Now the results are updated for PyTorch 1.12.1 and Branchformer (they got a...

Hi @siddhu001 > I would be interested to see if the postencoder is required to get the performance improvements. It looks like the postencoder gives some extra improvement: | PostEncoder...

Hi @sw005320 I would like to follow up on this. Can it be merged or should I change anything? I also have same setup for more SLU datasets: SLUE-VoxPopuli, CATSLU,...

@sw005320 this should be ready for a review (the CI failures are not related to my changes). The recipe is based on LibriHeavy Small and Mistral-7B-Instruct-v0.2. The results actually show...

> Thanks for adding this new support! I left some comments. Please let me know if there is any misunderstanding. Thank you for your review! I addressed your comments. Please...