bandicot

Long speech plays to the end, and dictation runs the full 30 seconds

Speech past a few seconds now reaches the player, dictation no longer stops at 12 seconds, and a recording can run for hours.

Product

/speak and any agent that calls the voice model now play the whole file, however long, from any folder the agent works in. Dictation takes the full 30 seconds and needs no special mode, because the voice models come with the app.

Agents can hand the speech and speaker models a recording of about four hours, up from about 13 minutes.

The model that comes with the app no longer stops an agent's scripts on a rough guess, so /transcribe reaches the models. Fixed safety rules decide instead.