# Local Whisper transcription

Nova can now transcribe local audio files with `whisper.cpp`.

## Current status

- `whisper.cpp` was built locally at `/tmp/whisper.cpp`
- verified working with the `tiny.en` model
- manual local transcription is ready
- Telegram voice-note wiring is still the next step

## Manual usage

```bash
chmod +x /data/.openclaw/workspace/scripts/transcribe_local_whisper.sh
/data/.openclaw/workspace/scripts/transcribe_local_whisper.sh /path/to/audio.wav
```

Optional flags:

```bash
/data/.openclaw/workspace/scripts/transcribe_local_whisper.sh /path/to/audio.wav --out /tmp/transcript.txt
/data/.openclaw/workspace/scripts/transcribe_local_whisper.sh /path/to/audio.wav --model tiny.en --language en --threads 2
```

## Important limitation for now

This script currently assumes a `whisper.cpp`-compatible local file, ideally WAV.
Telegram voice notes usually arrive as OGG/M4A, so the next step is to add a conversion/download pipeline before automatic Telegram transcription.