Every line below was sent to a text-to-speech provider twice: as the app wrote it, and after speechready.prepare. Both recordings were transcribed back to text. A misread is a transcript that differs from the intended words by more than a fifth.
| Provider | Model | Locale | Lines | Misread as written | Misread with speechready |
|---|
Three of the clearest fixes per voice. Press play on both sides. The transcript under each player is what a speech-to-text model heard.