Google Transcribe Open Issues

[pragmatrix]

Billing

• Billing records aren't supported yet.

Conversation Session

• With short models, single Utterance detection currently restarts, but I think we should just close the conversation in this case.
• The automatic restarting after the 305 seconds timeout works, but the detection does not work afterwards. It seems that there is an issue with the audio stream.

Model Specific

• The model latest_long seems to get stuck on single words it does not understand ... use combined fantasy words in German (need to check if somehow the final even may be suppressed if it's empty first).
• The chirp_3 model does not seem to support interim events.
• The long and short models (and perhaps the telephony model, too) do not support language detection as it seems. In these situations, a multi-language configuration should be rejected.
• Speaker diarization does not seem to be supported for Streaming Transcription at all, even though documentation states otherwise.