Shinji Watanabe

Thanks a lot. I recommend you make an ngram language model that is the same vocabulary as the ASR model. You can process the librispeech text with the ASR BPE...

I think so! Glad to hear that you successfully make it work.

This requires a high-level task/template design consideration. It also changes some basic ASR tasks and templates. Can you first summarize what this new function could do and what kind of...

@LiChenda, please reflect https://github.com/espnet/espnet/pull/2212 to `enh.sh`

@kamo-naoyuki, thanks for taking care of it!

Sure, we're definitely interested in it. Thanks for reaching us! Can you contact me by email ([email protected])?

> I'll be happy to integrate v2 version of our paper: https://arxiv.org/abs/2006.07637 Sounds good. We're now further refactoring our code and adding several examples. So, you can wait for our...

Thanks! We (me, @LiChenda, @Emrys365, and @shincling) will discuss how we can integrate your efforts.

FYI, @kan-bayashi, it is still >2.0s https://github.com/espnet/espnet/runs/5158915826?check_suite_focus=true#step:8:334 If it happens often, we may need to reduce the time more.

I suggest you cut the audio by using vad. We do not expect such long audio inputs.