voice-to-text-configlisted
Install: claude install-skill DmitriyYukhanov/claude-plugins
# Voice-to-Text Configuration
Set up local Whisper-based transcription for Telegram voice messages.
## Steps
Run each step sequentially. Report status clearly after each one.
### 1. Check faster-whisper installation
```bash
python -c "import faster_whisper; print(f'faster-whisper {faster_whisper.__version__} installed')" 2>&1
```
- **If installed**: print the version, move to step 2.
- **If ImportError**: tell the user and install it:
```bash
pip install faster-whisper
```
Verify the install succeeded before continuing.
### 2. Check / download Whisper model
The default model is controlled by the `WHISPER_MODEL` env var (default: `base`).
Available sizes: `tiny` (~40MB, fastest), `base` (~75MB, good balance), `small` (~250MB), `medium` (~750MB, most accurate for CPU).
Ask the user which model size they want if they haven't specified one. Then check if it's cached:
```bash
python -c "
import os, sys
from huggingface_hub import try_to_load_from_cache
model = os.environ.get('WHISPER_MODEL', 'base')
cached = try_to_load_from_cache(f'Systran/faster-whisper-{model}', 'model.bin')
if cached:
print(f'Model \"{model}\" is cached at: {cached}')
else:
print(f'Model \"{model}\" is NOT cached yet — needs download')
sys.exit(1)
" 2>&1
```
If not cached, download it:
```bash
python -c "
import os, sys
model = os.environ.get('WHISPER_MODEL', 'base')
print(f'Downloading whisper model \"{model}\"... (this may take a minute)')
from faster_whisper import WhisperMod