Audio chunk too short #191
Description
Hello,
I've recently used the openai-whisper cookbook example to add the Speech-to-Text function to my app.
I've noticed that the audio gets transcribed in small chunks instead of waiting for the audio recording to be stopped by the user.
This generates a low-quality transcription because each small audio chunk gets transcribed individually, sometimes it transcribes only half of the sentence and doesn't complete the transcription of the rest (see attached image).
Is there a way to set it in a way where the audio gets fully captured and the transcription starts only when the user manually stops the recording?
Hypothetically, I wouldn't mind if the recording stopped automatically after a few seconds of silence, but in this moment it looks like this function doesn't work very well - I tried upping the SILENCE_TIMEOUT to 4000.0 without much success-, so I'd rather do it manually.
Many thanks in advance.