We do not check the result status at all. Or rather, it seems like Google will try to transcribe as much of the audio chunks as possible until one errors.
The currently thinking about this is that Google will try to transcribe as many chunks as possible. BUT, if you reach a budget limit, it will stop, mid transcribe. Thus chunks before this point had proper transcriptions and chunks after this point (I assume) have error messages.
We should look for error messages before checking the transcript.
Google Speech-to-Text SR model simply assumes that the call to Google Speech-to-Text will run and return correctly: https://github.com/CouncilDataProject/cdp-backend/blob/main/cdp_backend/sr_models/google_cloud_sr_model.py#L151
We do not check the result status at all. Or rather, it seems like Google will try to transcribe as much of the audio chunks as possible until one errors.
An example of one chunk erroring is here: https://github.com/OpenMontana/missoula-council-data-project/runs/7758558858?check_suite_focus=true#step:8:190
Where it raises an
AttributeError.The currently thinking about this is that Google will try to transcribe as many chunks as possible. BUT, if you reach a budget limit, it will stop, mid transcribe. Thus chunks before this point had proper transcriptions and chunks after this point (I assume) have error messages.
We should look for error messages before checking the transcript.