How to apply this guide
Use consistent audio and settings to identify which compute route is faster and reliable on your device.
Start with
Use the same model, language, file and transcription settings. Compare first-run and repeated timings separately. Note the hardware and app or CLI version.
Check before you finish
Compare the final phrase and any names. Include the time spent opening the result and correcting it. Do not extrapolate one device’s result to every Mac.
Evidence to keep: Keep the target field open and verify that the complete reviewed result reached the intended destination. This page’s key limit is: Use consistent audio and settings to identify which compute route is faster and reliable on your device.
Start here
A workflow using Whisper locally may support CPU or GPU acceleration depending on your build and hardware. Compare actual completed results before inferring speed from the engine name.
Keep the jobs equivalent
Use the same model, language, file and transcription settings. Compare first-run and repeated timings separately. Note the hardware and app or CLI version.
Check whether the tool fell back
Read the log or engine indicator if the tool provides it. A successful run can follow a fallback after acceleration fails, so separate the reported compute route from the one actually used.
Use time to usable text
Compare the final phrase and any names. Include the time spent opening the result and correcting it. Do not extrapolate one device’s result to every Mac.




