MEASURE ON THIS DEVICE

Where does the generation time go?

This comparison uses your 20-minute swim with Emma announcing elapsed time every 30 seconds: 39 announcements, from 30 to 1170 seconds.

It runs GPU and CPU speech generation, each followed by a cached run, and encodes the complete recording each time. It does not edit your workout. First-time model downloads can increase loading time. A full comparison may take several minutes.

“Uncached speech” forces fresh narration; downloaded model files may already be cached. “Cached” reads saved narration using a fresh controller. GPU tests report unavailable/failed hardware rather than silently measuring CPU fallback. Keep this page open while it runs.

Ready to compare

No measurements yet.