Compare models with your voice

The best model depends on your voice, accent and microphone. This tool measures it on you.

Run it

1. Download at least one local model (two or more to compare). Open Settings → Speech → Local models.
2. Click Start… next to “Compare models with my voice”.
3. Read the five sentences, one at a time and at a normal pace: click Record, read, click Stop. You can redo any of them.
4. After three or more, click Compare.

Each downloaded model that covers your language recognizes the same recordings.

Reading the result

• Character error rate is the share of characters that are wrong, missing or extra, ignoring punctuation. Lower is better.
• The model with the lowest rate is marked Most accurate. If two models differ by less than 1%, none is marked, because that is within the noise of a few sentences.
• Average time per sentence shows relative speed.
• A model that writes no punctuation is flagged, so a good accuracy score does not hide that you would add punctuation yourself.
• Use this one switches to that model.
• “By sentence” shows exactly what each model heard.

Privacy

Recordings are kept in memory only. They are not written to disk and not uploaded, and they are cleared when you close the window. Cloud services are not part of this comparison.