How real-time transcription work?

[tuotuoshao]

I want to understand the differences between streaming and no-streaming transcription.

I have found out the clients and server all provide asynchronous function in streaming, and the clients truncate the audio data with a fixed time interval, and the whisper model is same with no-streaming(I have realized in some streaming asr projects, they use a context model, this may solve my confusion).

What troubles me is that when two truncated audio chunks all include a pronunciation of an character, or the pronunciation durations of this character are all too small, will the result be repeated or error?