[FEAT]: In-Browser Voice Recorder with Live Whisper Transcription

[krrishrastogi05]

📝 Description

FireForm supports voice memos as input but currently requires the user to record audio externally and upload a file. For field use — a firefighter finishing a shift in a truck — this friction is unacceptable. We should capture audio directly in the browser with live transcription displayed as they speak.

💡 Rationale

Eliminating the "record externally → find the file → upload" flow is the single biggest UX improvement for field responders. The MediaRecorder API is available in all modern browsers and requires no native app. Combined with a local Whisper endpoint (already in the FireForm stack via Ollama), this can be done entirely on-device with no cloud dependency.

🛠️ Proposed Solution

Frontend:

• Use the browser MediaRecorder API to capture microphone audio in-browser
• Stream audio chunks to the backend via WebSocket or chunked POST
• Display live transcription text as Whisper processes each chunk

Backend:

• Add a /api/v1/transcribe/stream WebSocket endpoint
• Pipe incoming audio chunks to the local Whisper model
• Return partial transcription tokens in real time

✅ Acceptance Criteria

• User can record audio directly in the browser without any file upload step
• Live transcription text appears on screen while recording
• Final transcription is passed into the existing LLM extraction pipeline seamlessly
• Works in Chrome and Firefox
• Graceful fallback to file upload if MediaRecorder is unavailable
• Works within the existing Docker container setup