enhancement: record audio and video from webcam simultaneously

[jeeanribeiro]

Description

From @MandarGogate issue:

Is your feature request related to a problem? Please describe.
I want to build a demo for machine learning model that processes Audio-Visual data from webcam. The current Video component records from webcam without audio.
Describe the solution you'd like
Since the Video component uses MediaRecorder API it can record audio and video simultaneously.

Changes:

- Enables audio in Webcam.svelte inside `access_webcam` function

Closes: #2704

Checklist:

• I have performed a review of the code
• I have added a short summary of my change to the CHANGELOG.md
• The code follows the style guidelines of this project

[jeeanribeiro]

@MandarGogate
I think that for this PR to be mergeable, it just needs the summary of the changes added in CHANGELOG.md, you can check the instructions about it here: https://github.com/gradio-app/gradio/blob/main/.github/PULL_REQUEST_TEMPLATE.md
As I'm trying to merge your fork, I don't have the permission to do this change

@abidlabs
Waiting for your review 😄

[pngwn]

We should probably make this an option to ensure the author has choice and that users aren't bombarded with permission requests when they run apps. Many models don't need or can't use both audio and video inputs.

[abidlabs]

Agreed with @pngwn‘s suggestion — ideally we have a Boolean parameter e.g. “include_audio” which can be True by default and applies to the webcam scenario but also to the regular Video component. In other words, if it is False then the audio from an uploaded Video would be removed as well

[MandarGogate]

I have added include_audio parameter. You can try it yourself using https://gist.github.com/MandarGogate/d49d1fd871b85c9840d5df7656c10c41

[pngwn]

Looks good to me. Left a few suggestions for the wording of the change log and docstring.

[pngwn]

Is defaulting to True a breaking change? Are there any times when the presence of an audio track will cause the predict function/ model that is processing the video to choke + fail?

[abidlabs]

Is defaulting to True a breaking change? Are there any times when the presence of an audio track will cause the predict function/ model that is processing the video to choke + fail?

Good point, let's test with a few Spaces that currently take in a webcam video. I believe @nateraw has built some...

[abidlabs]

Sorry for the late response @jeeanribeiro. I tested this out and it turns out setting include_audio to be True by default will be a breaking change because some users have webcam -> video demos where they do not expect the webcam to be recording the video. My suggestion would be to have include_audio be None, which gets set to True by default if source is upload but False if source is webcam. Would you be able to make that change and merge in the latest changes from main. Happy to re-review

cc @pngwn @aliabid94

[MandarGogate]

@abidlabs I have updated the default behaviour of include_audio and merged latest changes from main

[abidlabs]

Tested and works great. Couple of tests need to be fixed but I can go ahead and do that -- thanks for this addition to the gradio library @jeeanribeiro!