Deepgram Flux accepts mid-call ASR reconfiguration without reconnecting
- Source
- x.com
- Date
Most voice agents configure ASR once at connection time and live with it for the entire call. The call changes. The config doesn’t. We just fixed that. 🧵
Flux now supports on-the-fly configuration. Send a `Configure` message mid-stream to update keyterms and turn detection thresholds. No reconnecting. No parallel streams. Changes apply immediately.
Collecting a PIN? Make turn detection more patient so the user isn’t cut off. Moving to medication review? Swap in drug names. Back to open conversation? Tighten thresholds. One connection that shifts context as the call evolves.
Full details, code examples, and configurable parameters in the blog post: https://t.co/dRkNC0RB1H
Context
Deepgram's Flux can change its speech-recognition settings during an ongoing call. An application sends a Configure message over the existing connection to update key terms (words the recognizer should favor) and the thresholds that decide when someone has finished speaking. The technical announcement says the feature works without disconnecting or reconnecting, and that updates apply to the audio that follows.
This lets the application adapt to the current task. Deepgram's examples include loosening end-of-turn thresholds while a caller reads out a password or one-time code so they are not cut off, and swapping in medication names during a pharmacy conversation. The application chooses those changes; Flux returns a success or failure message so the application can check whether the requested configuration took effect.
- streaming — Sending a model's response token by token as it is generated, so the reader sees text immediately instead of waiting for the whole answer.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
Checking sign-in…
Loading comments…