Recipes for release

[pzelasko]

I'm thinking what are the recipes we can show at the tutorial, and whether we need to add more. AFAIK we have the following ready (and can use roughly any combination of them):

Corpora

• LibriSpeech
• Aishell

Architectures

• tdnn-lstm
• conformer
• contextnet

Topologies

• CTC

Criterions

• CTC
• MMI 2-gram dens

AM targets

• phones + blank
• BPEs + blank

[pzelasko]

I am thinking about adding a recipe for something conversational, to illustrate the need for/benefits of multiple supervision segments with actual acoustic context. I could prepare a recipe for this, e.g. on Switchboard? Not sure if there's a similar corpus that is also open (maybe GigaSpeech, possibly a subset of it).

[danpovey]

Cool! Yes, those features look good. And also your thing about multiple segments, although my focus right now is to simply have something -- anything-- ready by the time we give the Interspeech tutorial. Dan

…

On Sun, Jul 4, 2021 at 6:31 AM Piotr Żelasko ***@***.***> wrote: I am thinking about adding a recipe for something conversational, to illustrate the need for/benefits of multiple supervision segments with actual acoustic context. I could prepare a recipe for this, e.g. on Switchboard? Not sure if there's a similar corpus that is also open (maybe GigaSpeech, possibly a subset of it). — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#2 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAZFLO25FP64K2KBNMEL4EDTV6FTNANCNFSM47YRLOOQ> .

[pzelasko]

Don't worry, we'll make it ;)
Do you think we should copy the code from snowfall and adjust it somehow or just work on the snowfall repo for the time being? I think given that we're trying to limit the scope, maybe it makes sense to just stick to snowfall for now. Do you envision any serious changes to how the code is organized there by the tutorial deadline?

[danpovey]

I think we should take the opportunity to clean up interfaces a little and rework the directory structure a bit,
and get rid of any old versions of code that we don't want. But we don't need to make very fundamental changes, I think.