ASR Model Management
View, download, and manage ASR models. Models are stored locally on your Mac and can be re-downloaded at any time.
Available Models
| Backend | Model | Size | Notes |
|---|---|---|---|
| Qwen3-ASR | 0.6B | ~600 MB | Fast, lightweight. Best for Chinese + English. |
| Qwen3-ASR | 1.7B | ~1.7 GB | Higher accuracy. Recommended for most users. |
| MLX Whisper | Small | ~500 MB | Apple Silicon only. Good balance of speed and accuracy. |
| MLX Whisper | Large Turbo | ~1.6 GB | Apple Silicon only. Fast with high accuracy. |
| MLX Whisper | Large v3 | ~3 GB | Apple Silicon only. Highest accuracy. |
| Whisper | Tiny | ~75 MB | Very fast, lower accuracy. Good for testing. |
| Whisper | Base | ~140 MB | Fast with decent accuracy. |
| Whisper | Small | ~460 MB | Good balance for Intel Macs. |
| Whisper | Medium | ~1.5 GB | High accuracy, slower on Intel. |
| Whisper | Large | ~3 GB | Highest accuracy, requires 16 GB RAM. |
Downloading Models
Click the download button next to any model in Settings → General. A progress bar shows the download status in real time. You can continue using Echosy while a model downloads. Once downloaded, switch to it from the model dropdown in the recording bar.
Deleting Models
Click the delete button next to a downloaded model to remove it and free up disk space. You can always re-download it later. The currently active model cannot be deleted — switch to a different model first.
MLX Whisper models are optimized for Apple Silicon and use the GPU for fast inference. If you have an Apple Silicon Mac (M1/M2/M3/M4), these are recommended for the best Whisper performance.