Whisper Settings
Configure speech-to-text functionality using OpenAI's Whisper model.
Enable Whisper
Master toggle for voice input features.
When enabled:
- Voice input button appears in the chat interface
- Use the keyboard shortcut (default:
Cmd/Ctrl+Shift+V) to toggle voice input - Speech is transcribed locally using the Whisper model
Model Selection
Selected Model
Choose which downloaded Whisper model to use for transcription.
TIP
You must download at least one model before you can use voice input.
Language
Select the language for transcription:
- Auto Detect: Automatically detect the spoken language
- Or choose a specific language for better accuracy
Supported languages include: English, Chinese, Japanese, Korean, German, French, Spanish, Portuguese, Russian, Italian, and many more.
Available Models
Download Whisper models based on your needs:
| Model | Size | Speed | Accuracy | Best For |
|---|---|---|---|---|
| Large V3 Turbo (Recommended) | 1.6 GB | Fast | High | Best overall choice |
| Tiny | 75 MB | Fastest | Low | Quick tests, limited storage |
| Tiny (English) | 75 MB | Fastest | Low | English only, fastest |
| Base | 142 MB | Very Fast | Medium | Daily use, balanced |
| Base (English) | 142 MB | Very Fast | Medium | English only |
| Small | 466 MB | Fast | Good | Most users |
| Small (English) | 466 MB | Fast | Good | English only |
| Medium | 1.5 GB | Medium | High | High accuracy needs |
| Medium (English) | 1.5 GB | Medium | High | English only |
| Large V3 | 3.1 GB | Slow | Highest | Best accuracy |
English-Only Models
Models with "(English)" suffix are optimized for English and may provide better accuracy for English speech, but cannot transcribe other languages.
Managing Models
Downloading a Model
- Find the model you want in the Available Models list
- Click the Download button
- Wait for the download to complete
- The model will show a checkmark when ready
Download progress shows:
- Percentage complete
- Download speed
- Estimated time remaining
Deleting a Model
- Click the delete (trash) icon next to a downloaded model
- Confirm the deletion
WARNING
If you delete the currently selected model, Alma will switch to another available model or disable Whisper if no models remain.
Usage
Once configured:
- Click the microphone icon in the chat input, or use the keyboard shortcut
- Speak clearly into your microphone
- Your speech will be transcribed and inserted into the input field
- Edit if needed, then send your message