Live transcription
Every word, the moment it's spoken.
MdB Voice Assist transcribes caller speech in real time using Deepgram's Nova model. Audio is streamed over WebSocket as it arrives from the PSTN, so there is no wait between the caller speaking and the system understanding them.
Endpointing and voice activity detection are handled server-side. The transcript is passed to the reasoning layer only when a complete utterance has been detected — keeping the conversation natural and avoiding premature interruptions.