[Discussion - Improvements] - Real-time (or near-real-time) transcription in the browser with React

[avie41]

Hello @ggerganov

I managed to reuse the code from the stream example and integrate it into a React application using Vite.js.

Keeping the basic implementation, adapted in TypeScript, I have a latency of about 1.5, 2 seconds on average.

But it looks like the implementation given in the example presents a fairly basic audio chunking strategy that could be improved.

• Has any work already been done on this?
• Could the CPP code that is then compiled with Emscripten be improved?

Additional context:

At the moment, my application uses vosk-browser, which plugs into an Audio streamer. I would like to turn to Whisper for its superior transcription quality and would like to optimize my implementation as much as possible to get closer to realtime with whisper.cpp.

[qxprakash]

hi @avie41 can you share the code of your implementation , I wanted to get streaming to work in the whisper.cpp server.