I am working on remote transcription using Watson STT. I stream audio from a mic to an mqtt client that interfaces with watson stt via websocket.
Everything words fine if I follow this flow.
RECORD CHUNK -> RECORD CHUNK -> RECORD CHUNK -> PUBLISH CHUNK -> PUBLISH CHUNK ->PUBLISH CHUNK
The only problem is that I have to wait until I finish recording to start transcribing, which is kind of slow.
To speed things up, I want to have the recognition happen while I am recording. So I followed this
RECORD CHUNK -> PUBLISH CHUNK ->RECORD CHUNK -> PUBLISH CHUNK ->RECORD CHUNK -> PUBLISH CHUNK ->RECORD CHUNK
When I do this I never get a final transcript from Watson STT. The only thing I can think of is that the latency between chunks is increased and Watson does not tolerate that?