Which Whisper model should I use?
The right answer is the largest model your machine can hold without swapping, adjusted for how much you care about speed. Two questions and you have it, plus the full size and speed table so you can overrule the recommendation.
Runs entirely in your browser. There is nothing here to send anywhere.
Use this one
small
The usual answer. Noticeably better than base on names and technical words, and still much faster than realtime.
466 MB on disk, about 1.0 GB while it runs, roughly 6× the speed of large.
| Model | Download | Memory | Speed |
|---|---|---|---|
| tiny | 75 MB | 390 MB | 32× large |
| base | 142 MB | 500 MB | 16× large |
| small | 466 MB | 1.0 GB | 6× large |
| medium | 1.5 GB | 2.5 GB | 2× large |
| large | 2.8 GB | 4.6 GB | 1× large |
Sizes are the ggml weights whisper.cpp downloads. Speed is relative to the large model on the same hardware, so read it as an order of magnitude rather than a benchmark for your machine.
Local recording tips, once in a while
Short notes on recording and transcribing meetings without sending them anywhere. Unsubscribe whenever.
Stop doing this by hand
autorec records your Zoom, Teams and Meet calls on your own machine and transcribes them locally. No bot joins the call, and nothing is uploaded. €20 once.
Download autorec