Submit ProductSign in
A small speech-to-text model that writes words before they end
Grounded in available product and source data
Earlybird is a streaming speech-to-text model that puts a word on screen while the speaker is still saying it. Each word comes with a confidence score, so you can see how sure the model is before you keep or discard the line.
The model is small enough to run locally and was trained on audio cut off at random points, including partway through a word. On audiobook speech a typical word lands well before it ends, which gives voice agents time to start a search or fetch a record during the conversation.
