Live transcription - stream text output while recording

[slonka]

Description

Support live/streaming transcription that outputs text in real-time as the user speaks, rather than waiting until recording stops to process the entire audio at once.

I'm happy to take a stab at this if this is something that you'd want to see in the project.

Current behavior

1. User starts recording
2. User speaks
3. User stops recording
4. Audio is sent to whisper.cpp / OpenAI API for processing
5. Transcribed text is pasted

Proposed behavior

1. User starts recording
2. As the user speaks, transcribed text appears in real-time
3. User stops recording
4. Final text is pasted (or incrementally inserted as it's produced)

Notes

• whisper.cpp supports streaming mode via --stream flag
• OpenAI Whisper API does not currently support streaming
• sherpa-onnx supports streaming ASR models natively
• This would pair well with VAD (Voice Activity Detection (VAD) #185) for detecting speech segments