Open benrito opened 7 months ago
What's happening is that we grab is_final but keep transcribing, because endpointing is lower than utterance final threshold: https://developers.deepgram.com/docs/understanding-end-of-speech-detection
Then, we take the second is_final and ignore the first one.
We need to do testing on the best logic for end of speech detection, and then make sure that no matter what, we take the first is_final only (or else ensure that the whole transcript is passed as is_final)
Investigate whether sample_rate is causing problems or needs to be hardcoded according to device Socket timeout