Alexis
English Flux TTS
Flux TTS: More context. Better conversation.
Give every turn the context of the full conversation with Deepgram Flux TTS, text-to-speech built for real-time conversations.
Try for freeLeading voice teams build with Deepgram
The voices for your voice agents
Expressive by default, fast under load, deployable anywhere, and accurate where it counts most.
Expressive by default
Empathetic when the customer is frustrated, precise when details matter, and natural across every turn. No SSML, style tags, or prompt engineering required.
Turn-taking, built in
Persistent state, native interruption handling, and turn lifecycle live inside the API. Ship voice agents without writing the orchestration yourself.
Compliance without compromise
Run Flux TTS in your own cloud, VPC, or on-prem to meet data residency, security, and compliance requirements.
Responses in as low as 80ms
Start producing audio in as little as 80 milliseconds.. Audio is expressive and fast, no tradeoffs.
Lowest error rate tested
Drug names, account numbers, and alphanumerics handled with ease.
Real-time control
Change speed, pronunciation, or style mid-conversation without restarting. The server handles chunking so you don't have to.
Choose a voice, hear it live
Explore voices
Drew
English Flux TTS
Haley
English Flux TTS
Jack
English Flux TTS
More accurate where it counts
Traditional TTS stumbles on drug names, account numbers, tracking numbers, and IVR strings. Flux TTS has it handled.
- 2.2%Median word error rate
- 80msTime to first audio, as low as
- 3XLower WER on hard prompts
Adaptable (and accurate) across industries
Healthcare
Warm at intake, precise at handoff
HIPAA-ready voice agents that handle intake, scheduling, and reminders through natural conversation.
See healthcare: Warm at intake, precise at handoff
Government
Every mission runs on voice
Deploy in GovCloud, on-prem, or fully air-gapped. We never train our models on government audio without your consent.
See government: Every mission runs on voice
Restaurants
Loud kitchens, clear orders
Voice AI ordering, employee assist, and reservations for restaurant operations, and fully integrated with your POS.
See restaurants: Loud kitchens, clear orders
Contact centers
Empathy at enterprise scale
Natural, human-like voice for real-time conversational AI, scaled. Handles cross-talk, background noise, and the alphanumerics.
See contact centers: Empathy at enterprise scale
Related articles
Yael Scarlett
Introducing Flux TTS: Conversation-native text to speech for real time voice agents
Announcements
Deepgram
Emotional text-to-speech: how expressivity controls fix flat delivery
AI Engineering & Research
Deepgram
Why TTS latency grows with response length and the architecture that stops it
AI Engineering & Research
Deepgram
Build a voice agent with Pipecat and Deepgram Flux STT and Flux TTS
AI Engineering & Research
FAQs
Deepgram Flux TTS is purpose-built for real-time voice agents. It reads the whole conversation instead of one line at a time, handles interruptions natively, and delivers expressive voices with time to first audio as low as 80ms. In blind head-to-head listening tests across 12 models from nine vendors, Flux TTS ranked first on both naturalness and expressiveness — the only model tested to top both.
ElevenLabs is built for expressive voice generation across studio, content, and agent use cases, with expressiveness controlled through manual audio tags. Flux TTS is built specifically for real-time voice agents, with tone and pacing chosen automatically from conversation context. In benchmark testing, Flux TTS had roughly half the word error rate of ElevenLabs v3 on read-aloud prompts.
Cartesia Sonic is known for low-latency voice generation. Flux TTS matches Cartesia on speed while adding conversation-native architecture: persistent state across turns, native interruption handling, and expressive delivery from context. In benchmark testing, Flux TTS had roughly one third the word error rate of Cartesia Sonic 3.5 on read-aloud prompts.
Deepgram Flux TTS is a conversation-native text-to-speech model built for real-time voice agents. It reads the whole conversation, not just the current line, so tone, pacing, and emotional register persist across every turn. Flux TTS delivers expressive voices with time to first audio as low as 80ms and native interrupt handling.
When a caller barges in, the Flux TTS server reports what the user actually heard, including the text spoken, text remaining, and words spoken. The agent resumes cleanly from the correct point instead of guessing or repeating. Interruption handling lives inside the API, not in orchestration code developers have to write.
Flux TTS is purpose-built for real-time voice agents, with conversation-native architecture, native interruption handling, and expressive delivery from context. Deepgram Aura-2 delivers efficient text-to-speech for production at scale. Clear, dependable speech for teams that prioritize cost efficiency, reliability, and high-volume deployment.