Alexis
English Flux TTS
Fast, expressive text-to-speech. Finally.
TTS built for voice agents that streams in as low as 80ms, stays expressive turn after turn, and deploys wherever your data lives.
Leading voice teams build with Deepgram
Speaking of features
Tone that holds up, wicked fast, and accurate on the hard stuff.
Conversationally aware
Tone, pacing, and emotional register carry across every turn, with no SSML, style tags, or prompt engineering required.
Fast and expressive
Responses start in as low as 80ms at full 24kHz fidelity, even under production load. Speed and expressive range in the same voice.
Maintains accuracy
Lowest word error rate of any TTS tested. Handles drug names, alphanumerics, account numbers, and dates with ease.
Native interruption handling
When a caller barges in, the server reports what was heard. The agent resumes cleanly, instead of guessing.
Flexible deployment
Run in the cloud, your private cloud, or on-prem. Get full control for regulated workloads in healthcare, finance, and government.
Predictable pricing
Starting at $0.045 per 1,000 characters, with volume pricing as low as $0.030.
Choose how you sound
Pick from natural, expressive voices built for real-time conversations.
Explore voices
Drew
English Flux TTS
Haley
English Flux TTS
Jack
English Flux TTS
One model family for speaking and listening
Deepgram Flux TTS handles speaking, Flux STT handles listening. Both run on the same conversation-native foundation instead of a patchwork of vendors, so tone and state always carry through.
- 80msTime to first audio, as low as
- 2.2%Word error rate, lowest of any TTS tested
- 24kHzNeural-codec audio fidelity
Built for different kinds of voice projects
From real-time voice agents to speech at enterprise scale, find the right model for the job.
Flux TTS
Conversational text-to-speech that carries tone across turns and handles interruptions natively.
- Expressive
- Context-aware
- Real-time
Aura-2
Efficient text-to-speech at scale for teams that need high-volume deployment.
- Cost-efficient
- Reliable
- Production-ready
Related articles
Yael Scarlett
Introducing Flux TTS: Conversation-native text to speech for real time voice agents
Announcements
Deepgram
The rise of PER: A new standard for measuring TTS accuracy
AI Engineering & Research
Jose Nicholas Francisco
How to evaluate voice quality claims across text-to-speech providers
AI Engineering & Research
Deepgram
TTS with emotion: emphasis and control in live conversations
AI Engineering & Research
FAQs
Text-to-speech (TTS) is the technology that turns written text into spoken audio. Modern TTS models generate natural-sounding voices for use cases like voice agents, IVR, audiobooks, accessibility, and any product where a computer needs to speak.
Deepgram's Flux TTS reads the whole conversation, not just the current line. Tone, pacing, and pronunciation carry across turns. Responses start in as low as 80ms. Interruptions, turn boundaries, and state are handled by the model, not custom code.
ElevenLabs is built for content generation and studio performance, with per-utterance expressiveness controlled by manual tags. Deepgram’s Flux TTS is built for live voice agents, with automatic expressive delivery across turns, streaming as low as 80ms, and flexible deployment for regulated workloads.
Cartesia and Flux TTS both target real-time voice. Deepgram’s Flux TTS model has a 3x lower word error rate on hard prompts, a full conversational protocol with native interruption handling, and pairs directly with Deepgram STT for a shared listening and speaking foundation.
Yes. Flux TTS ships in English plus 10 additional languages with built-in text normalization: Spanish, Portuguese, French, Japanese, German, Dutch, Hindi, Italian, Russian, and Mandarin Chinese.
The same multilingual model also supports 40+ additional languages for teams that provide their own text-normalization layer. These include Arabic, Belarusian, Bengali, Bulgarian, Cantonese Chinese, Catalan, Croatian, Czech, Danish, Estonian, Filipino, Finnish, Greek, Gujarati, Hebrew, Hungarian, Indonesian, Kannada, Korean, Latvian, Lithuanian, Macedonian, Malay, Malayalam, Marathi, Norwegian, Persian, Polish, Romanian, Serbian, Slovak, Slovenian, Swedish, Tamil, Telugu, Thai, Turkish, Ukrainian, Urdu, and Vietnamese.
Voices remain consistent across turns within a conversation, with language-specific voices for each market.
Time to first audio under 300ms is the ceiling for conversation to feel natural. Flux TTS starts audio in as low as 80ms, even under production load, so the voice arrives before a caller notices a gap.
Yes. Deepgram supports HIPAA-compliant deployments. Self-hosted and private cloud options keep processing inside your environment, and BAAs are available for eligible customers.
Yes. Flux TTS supports self-hosted deployment for teams with data residency, security, or compliance requirements. Run in your own cloud, private cloud, or on-prem infrastructure.
Yes. Deepgram integrates with Twilio, LiveKit, Daily, Pipecat, jambonz, and other telephony and voice agent platforms. Voice agents built on Deepgram handle inbound and outbound calls at production volume.
Flux TTS is $0.045 per 1,000 characters on pay-as-you-go with significant volume discounts. Aura-2 is $0.030 per 1,000 characters and Aura-1 is $0.015 per 1,000 characters, both with additional volume discounts. Enterprise discount tiers are available across all models.
Customer support agents need a voice that stays warm across long calls, recovers from interruptions, and pronounces account details accurately. Flux TTS was designed for exactly this shape of conversation.
Match the voice to the moment. Empathetic tone for support. Precise delivery for technical or medical detail. Warmth for consumer flows. Try voices side by side at talk.deepgram.com before you ship.