Hello speech, we love you too.

Improvements to our speech recognition technology.

As voice interactions are becoming more prevalent, we know that speech is important to you. Speech has been essential to our mission since our early beginnings, and we remain dedicated to enabling you with quality speech recognition technology.

We have two exciting news to share:

1. We’ve updated our baseline speech recognition models for English, which improved accuracy and latency.
2. Our speech recognition training loop now takes into account all validated audio samples to improve our models. This works the same way newly validated NLU samples improve the models serving built-in entities. Every sample you validate (pictured below) will be taken into account in each subsequent re-training of the speech recognition models. While the text retraining is almost instantaneous, the speech models will be updated weekly.

And yes, more languages are coming. 🔥

This system ensures our speech models continuously improve, and that your time spent in correcting audio transcriptions helps refine the models for everyone. This is the power of community: giving and receiving back. ❤

For more information on how to use our speech endpoint, see the documentation for POST /speech.

Thank you for your feedback and your patience on this one. We hear you. Please keep the feedback coming here!