Daniel Galvez
Daniel Galvez
I agree on the filepaths with spaces thing. Just not worth quoting almost every single variable evaluation even though it's "right" to do. On Thu, Sep 14, 2017 at 7:06...
Wow, there's a bug in `find_bad_utts_nnet.sh` where chain models don't use the right frame subsampling factor because of a typo. (I discovered this independently, not from combing through shellcheck's warnings.)...
I just wanted to note that I am encountering the same issue. Thank you for the detective work and work around @BrettRyland !
I can probably implement RNN-T loss function, at some point, if there is enough interest. It is straight forward to implement if you have an understanding of the CTC loss...
Thanks for letting me know @jonathanasdf
Jonathan Shen is correct. I've used the docker container successfully without having to preprocess the librispeech data every time after exiting, since I used the -v option. On Mon, Aug...
> I like the version history idea because it helps to replicate experiments that were performed on an older version of the dataset. One problem with the history idea is...
> There are no PyTorch versions for cuda 12.3 and we don't have such an environment locally. Just to note, there is no problem with mixing cuda toolkit versions within...
I can take a look at some point, but it's not high priority for me right now. If someone else runs into this issue, they should make a comment here.
Not stale. I've been incorporating work done in this branch piecemeal into the main branch.