Deepgram

Deepgram offers a range of English-speaking voices for its text-to-speech API, each designed to produce natural-sounding speech output in an array of different accents and speaking styles.

Deepgram's voices are promised to have human-like tones, rhythm, and emotion, lower than 250 ms latency, and are optimized for high-throughput applications.

Consult Deepgram's TTS models guide for more information and samples for supported voices.

Voice IDs

Copy the voice ID from the Values column of Deepgram's Voice Selection reference. Prepend deepgram. and the string is ready for use. For example: deepgram.aura-athena-en

Models

Deepgram's Aura model provides ultra-low latency text-to-speech optimized for conversational AI:

Model	Description
aura	Default - Low-latency conversational TTS with human-like tones

All Deepgram voices use the Aura model architecture, delivering sub-250ms latency.

Voices

Deepgram Aura voices are designed for natural-sounding English speech. Each voice follows the pattern aura-<name>-en.

Popular voices include:

• aura-asteria-en - Clear, professional female voice
• aura-luna-en - Warm, conversational female voice
• aura-stella-en - Energetic, friendly female voice
• aura-athena-en - Authoritative female voice
• aura-hera-en - Mature, confident female voice
• aura-orion-en - Deep, authoritative male voice
• aura-arcas-en - Professional male voice
• aura-perseus-en - Friendly male voice
• aura-angus-en - Casual male voice
• aura-orpheus-en - Smooth, articulate male voice
• aura-helios-en - Energetic male voice
• aura-zeus-en - Powerful, commanding male voice

For a complete list, consult Deepgram's Voice Selection guide.

Configuration

Deepgram TTS requires a Deepgram API key:

• DEEPGRAM_KEY - Your Deepgram API key

Sign up at the Deepgram Console.

Voice IDs

Deepgram voice IDs conform to the following format:

deepgram.<voice>

Where <voice> is the voice identifier from Deepgram's Voice Selection reference.

Examples:

deepgram.aura-asteria-en
deepgram.aura-orion-en
deepgram.aura-luna-en
deepgram.aura-zeus-en

Note: All Deepgram voices are currently English-only and use the Aura model by default.

---

Examples

Learn how to use Deepgram voices on the SignalWire platform.

SWML

Use the languages SWML method to set one or more voices for an AI agent.

version: 1.0.0
sections:
  main:
  - ai:
      prompt:
        text: Have an open-ended conversation about flowers.
      languages:
        - name: English
          code: en-US
          voice: deepgram.aura-asteria-en

Alternatively, use the say_voice parameter of the play SWML method to select a voice for basic TTS.

version: 1.0.0
sections:
  main:
  - set:
      say_voice: "deepgram.aura-asteria-en"
  - play: "say:Greetings. This is the Asteria voice from Deepgram's Aura text-to-speech model."

RELAY Realtime SDK

// This example uses the Node.js SDK for SignalWire's RELAY Realtime API.
const playback = await call.playTTS({ 
    text: "Greetings. This is the Asteria voice from Deepgram's Aura text-to-speech model.",
    voice: "deepgram.aura-asteria-en",
});
await playback.ended();

Call Flow Builder

Deepgram voices are not yet supported in Call Flow Builder.

cXML

<?xml version="1.0" encoding="UTF-8"?>
<Response>
<Say voice="deepgram.aura-asteria-en">
    Greetings. This is the Asteria voice from Deepgram's Aura text-to-speech model.
</Say>
</Response>