Performance Tips
Optimize Speakly for your system
Local Model Performance
- Use the quantized turbo Q5 (
large-v3-turbo-q5_0) for faster results - On Apple Silicon macOS, tick Enable next to GPU acceleration (Metal) under Settings → Transcription; on Intel, Windows and Linux the control does not appear, because no acceleration is available
- Close other heavy applications during transcription
- Consider BYOK for faster processing on slower hardware
Memory Usage
- Larger models need more memory and more CPU time — if transcription is slow, drop one size before trying anything else
- Close unused tabs/applications if memory is limited