Simple audio tutorial

[dvisockas]

Hello,

Seems like pytorch-audio is an under-served part of pytorch in its tutorials page.

I'd like to contribute by writing an initial tutorial starting with a simple audio classification (VCTK or UrbanSound for starters).

What are the requirements for that? Are there any guidelines? Maybe there are people working on torch-audio I could ping if something is not clear?

[vincentqb]

Thanks for reaching out! Feel free to add "cc @vincentqb" if you want to ping my attention directly here :)

I'd like to contribute by writing an initial tutorial starting with a simple audio classification (VCTK or UrbanSound for starters).

Yes, that would be a very nice addition to the tutorials if it can illustrate some feature of torchaudio. I wrote the torchaudio tutorial in #572 by creating a jupyter notebook. I then did some adjusting on the final rst file directly.

What are the requirements for that? Are there any guidelines? Maybe there are people working on torch-audio I could ping if something is not clear?

I'd say a good tutorial should be fairly short to run, say less than ~2 minutes. It's not a hard stop, but, unless there is something exciting to look at while code is running, it may not keep people's attention as much as a shorter tutorial to run.

I'm myself looking into adding a tutorial inspired by the wav2letter pipeline in torchaudio. I'm also considering refreshing an older audio classification tutorial we had using one of the dataset in torchaudio, see #1204.

[vincentqb]

Before you get started, can you provide a little more details? What model and/or dataset did you have in mind? UrbanSound is not in torchaudio, would you like to add it in there first?

[carljparker]

I'm closing this for now. Feel free to re-open if this is something you'd like to pursue.