2026.10.09 Agent STT SaaS

Added

• Added support for 8kHz audio. The sample rate in audio format now accepts 8000 or 16000 (default 16000).

• ForceEndOfUtterance can now be sent without a timestamp in external turn detection mode. Without a timestamp the turn is cut at the end of the audio received so far. We recommend using a timestamp for accuracy

• The client now receives a Warning, once per session, if audio is arriving faster than real time. Sending audio faster than real time is not recommended, as it can reduce accuracy over prolonged periods.

Fixed

• ForceEndOfUtterance messages sent in vad turn detection mode now provide an unsupported message Warning back to the client. ForceEndOfUtterance is only supported in external turn detection mode.