Summary

• New POST /transcribe/stream SSE endpoint that emits stage-by-stage progress events during transcription
• Frontend reads the stream and displays the current stage label (e.g., "Detecting speech…", "Aligning words…") alongside the percentage
• The original POST /transcribe endpoint is preserved for backward compatibility

Before

The progress bar showed a generic "Transcribing... X%" with no percentage updates until completion — the user had no idea which stage the pipeline was in.

After

Real-time stage labels with approximate progress:

Implementation

Backend: The SSE endpoint runs transcription in a thread executor. A progress_cb(pct, label) callback is threaded through transcribe_audio() → _transcribe_whisperx(), emitting progress at each pipeline stage. Events are serialized as data: {json}\n\n SSE format.

Frontend: The transcribeVideo function reads the SSE stream with a ReadableStream reader, parsing each line for progress/label/result events.

Files changed

• backend/routers/transcribe.py — New /transcribe/stream endpoint with SSE streaming
• backend/services/transcription.py — Add progress_cb parameter, emit stage progress
• frontend/src/App.tsx — SSE reader in transcribeVideo(), transcriptionLabel state, updated progress display

Test plan

• Open a video file — progress display should show stage labels updating in real-time
• Verify the old /transcribe endpoint still works (backward compatibility)
• Check that cached transcriptions return immediately with "Done (cached)"
• Cancel/close during transcription — should not leave orphaned tasks