Deepgram Flux TTS

About

Flux TTS is Deepgram's text-to-speech model built for real-time voice agents. It conditions on the whole conversation rather than the current line, so pacing, emphasis, and emotional register follow context without SSML or style tags, and it runs with Flux STT as one stack on a single connection so listening and speaking are timed together. Deepgram reports it first in its own blind head-to-head listening test, with a 73.4% naturalness win rate across 12 models and roughly 14,400 paired comparisons.