Skip to main content
Deepgram's Docs

Search documentation

Type to search this documentation.

On this pageOverview

Flux TTS Overview

Flux TTS brings the Flux promise to speech synthesis: a model and API built for the realities of a voice agent pipeline, not a one-shot text-to-audio pipe. It is served on the /v2/speak endpoint and shares one set of Flux voices across two transports. When a user barges in, Interrupt reports exactly what they heard (text_spoken) — the hard part of agent state reconciliation, answered by the API instead of reconstructed by hand.

Transport Endpoint Best for
Real-time / conversational (WebSocket) wss://api.deepgram.com/v2/speak Live voice agents: stream LLM tokens in, stream audio back, interrupt and resume across turns
Batch / pre-recorded (REST) POST https://api.deepgram.com/v2/speak Pre-generating fixed audio (IVR prompts, notifications, audiobook lines) where the whole text is known up front

Not sure which to use? See Batch vs Streaming.

  • Streaming-first — Stream LLM tokens straight into the socket; the server places flush boundaries internally.
  • Turn-based lifecycle — Each agent response is a turn with a clean start/finish, reported per turn.
  • Cross-turn voice consistency — Conversational state persists across turns, so tone carries forward. See Cross-Turn Context.

On barge-in, Interrupt reports exactly what the user heard, and mid-stream Configure adjusts speed without reconnecting.


Suggest an edit

Propose a replacement for this page. The site team reviews it before applying any changes.

Export
Documentation menu