Extensions: Firebase extension for STT

[pietrop]

What feature would you like to see?

Similar to how there is an extension for Google Translate, I was wondering
if there are any plans to make an extension for Google Cloud Speech To Text?

I'd be happy to contribute to the open source development, if this gets picked up and there are some general directions on how to go about developing a firebase extensions etc..

I've already raised it through Firebase support directly, and they mentioned to raise it here as well.

How would you use it?

I'd use it to make Apps that have a transcription service component with Firebase, eg bbc/digital-paper-edit, autoedit2.

[Ehesp]

Thanks @pietrop

cc @karayu do we have an internal tracker for feature requests for discussion/implementation?

[Ehesp]

Hi @pietrop could you please add the feature request here: https://firebase.google.com/support/troubleshooter/report/features/ - linking this issue as reference. Thanks!

[pietrop]

Thanks @Ehesp , I've just submitted it

[pietrop]

👋 I am looking more closely into a possible implementations for this, so thought I'd share some notes for doing Cloud Function + STT - long running audio files

and some options I've considered

1. STT + Cloud function
   Cloud function times out 1min to 9min. So would not be able to get results for a long running audio file. But would work ok, for shorter files. Hard to estimate what's the max audio length it would support?

2. Cloud function - to start STT
   based on the docs, the cloud function could be used to start the STT, and store the name of the operation. This could be saved in firestore, and another function could be used to do some pooling, at a later stage to get the results when ready. Results are stored for 5 days.

3. Cloud Function + App Engine
   The Firebase cloud function could trigger an app engine instance, that could run the recognition. I think this would probably not time out like the cloud function, and could save the results to firestore when done.

[Ehesp]

But would work ok, for shorter files. Hard to estimate what's the max audio length it would support?

I'm not very knowledgeable on video files, however I also imagine there would be more to it than the length; quality etc may play a part.

Interesting discussion though to help consider scope.

[pietrop]

I think for getting STT of a video, you'd have to first convert it to audio, to meet the audio specs expected by GCP STT (eg see firebase/functions-samples/ffmpeg-convert-audio) and then send it to GCP STT.

In any case at the moment I am trying the second option (2. Cloud function - to start STT) with a google cloud tasks.

[jhuleatt]

Since we've recorded this request in the internal tracker, I'm closing out this issue.