Configure

TTS Models

An overview of Text-to-Speech providers and models you can use with the Voice Agent API.

By default Deepgram Text-to-Speech will be used with the Voice Agent API. Deepgram supports two text-to-speech model families, and the agent picks the right TTS endpoint based on the version field of agent.speak.provider — you do not manage endpoint URLs yourself.

  • Flux TTS (v2) for streaming-first, voice-agent-first synthesis with turn-based lifecycle and cross-turn voice consistency.
  • Aura (v1) for the broadest, generally available Deepgram voice catalog across English and Spanish.

You can also use Deepgram’s native Cartesia support or opt to use another provider’s TTS model with your Agent by applying the following settings.

You can set your Text-to-Speech model in the Settings Message for your Voice Agent. See the docs for more information.

Deepgram TTS models

Deepgram offers two TTS model families for the Voice Agent API. Set agent.speak.provider.version to v2 for Flux TTS or v1 for Aura. When you omit version, the agent defaults to v1 (Aura).

Choosing a model family

Flux TTS (V2)Aura (V1)
Best forstreaming-first voice agentsbroadest GA voice catalog
AvailabilityEarly AccessGenerally Available
Turn-based lifecycleyesno
Cross-turn voice consistencyyesno
provider.versionv2 (required)v1 (default)
Model stringflux-{voice}-{language} (e.g. flux-alexis-en)aura-2-thalia-en, aura-asteria-en, etc.

Flux TTS

Early Access. Flux TTS is in Early Access — the Flux TTS-specific API surface in the Voice Agent and voice catalog may change before general availability.

Flux TTS is Deepgram’s streaming-first, voice-agent-first text-to-speech model family. It brings a turn-based lifecycle and cross-turn voice consistency built for the realities of a voice agent pipeline. For details, see the Flux TTS Feature Overview.

ParameterTypeDescription
agent.speak.provider.typeStringMust be deepgram
agent.speak.provider.versionStringMust be v2
agent.speak.provider.modelStringFlux TTS model id in the format flux-{voice}-{language}, for example flux-alexis-en. See the Flux TTS voice catalog for the full list.

Flux TTS voices are served only on agent.speak.provider.version: v2, and Aura voices only on agent.speak.provider.version: v1. Switch families by changing agent.speak.provider.version and model together.

Example

JSON
1{
2  "agent": {
3    "speak": {
4      "provider": {
5        "type": "deepgram",
6        "version": "v2",
7        "model": "flux-alexis-en"
8      }
9    }
10  }
11}

Aura

For a complete list of Deepgram Aura TTS models see TTS Voice Selection.

ParameterTypeDescription
agent.speak.provider.typeStringMust be deepgram
agent.speak.provider.versionStringOptional. Defaults to v1 when omitted.
agent.speak.provider.modelStringThe Aura TTS model to use, for example aura-2-thalia-en.
agent.speak.provider.speedFloatSpeaking rate multiplier. Available for Aura (provider.version v1) only. Range: 0.7 - 1.5. Defaults to 1.0. See TTS voice controls for details.

Example

JSON
1{
2  "speak": {
3    "provider": {
4      "type": "deepgram",
5      "model": "aura-2-thalia-en",
6      "speed": 0.9
7    }
8  }
9}

Deepgram-managed Cartesia TTS models

Deepgram also provides managed support for Cartesia TTS. For a complete list of Cartesia TTS models, visit Cartesia’s TTS Docs. Cartesia is included in the Standard pricing tier.

ParameterTypeDescription
agent.speak.provider.typeStringMust be cartesia
agent.speak.provider.modelStringThe TTS model to use
agent.speak.provider.speedString or NumberSpeaking rate control. Accepts slowest, slow, normal, fast, fastest, or a numerical value for more granular control. See Cartesia speed documentation.

Example

JSON
1{
2"agent": {
3  "speak": {
4    "provider": {
5      "type": "cartesia",
6      "model_id": "sonic-2",
7      "voice": {
8        "mode": "id",
9        "id": "a167e0f3-df7e-4d52-a9c3-f949145efdab"
10      },
11      "speed": "normal"
12    }
13  }
14}
15}

BYO Third Party TTS models

To use a third party TTS voice, specify the TTS provider and required parameters.

OpenAI

For OpenAI you can refer to this article on how to find your voice ID.

ParameterTypeDescription
agent.speak.provider.typeStringMust be open_ai
agent.speak.provider.modelStringThe TTS model to use
agent.speak.provider.voiceStringThe voice to use
agent.speak.endpointObjectRequired and must include url and headers
agent.speak.endpoint.urlStringYour OpenAI API endpoint URL
agent.speak.endpoint.headersObjectRequired headers for authentication

Example

JSON
1{
2  "agent": {
3    "speak": {
4      "provider": {
5        "type": "open_ai",
6        "model": "tts-1",
7        "voice": "alloy"
8      },
9      "endpoint": {
10        "url": "https://api.openai.com/v1/audio/speech",
11        "headers": {
12          "authorization": "Bearer {{OPENAI_API_KEY}}"
13        }
14      }
15    }
16  }
17}

Eleven Labs

For ElevenLabs you can refer to this article on how to find your Voice ID or use their API to retrieve it. See their TTS Docs for more information. ElevenLabs does not support WebSocket streaming for the eleven_v3 model - instead, use the HTTPS REST endpoint (see example).

We support any of ElevenLabs’ Turbo 2.5 voices to ensure low latency interactions

ParameterTypeDescription
agent.speak.provider.typeStringMust be eleven_labs
agent.speak.provider.model_idStringThe model ID to use
agent.speak.provider.language_codeStringOptional Language code
agent.speak.endpointObjectMust include url and headers
agent.speak.endpoint.urlStringYour Eleven Labs API endpoint URL
agent.speak.endpoint.headersObjectHeaders for authentication

Example

JSON
1{
2  "agent": {
3    "speak": {
4      "provider": {
5        "type": "eleven_labs",
6        "model_id": "eleven_turbo_v2_5",
7        "language_code": "en-US"
8      },
9      "endpoint": {
10        "url": "wss://api.elevenlabs.io/v1/text-to-speech/{voice_id}/multi-stream-input",
11        "headers": {
12          "xi-api-key": "{{ELEVEN_LABS_API_KEY}}"
13        }
14      }
15    }
16  }
17}

Example (eleven_v3 via HTTPS)

Because eleven_v3 does not support WebSocket streaming, use the HTTPS REST endpoint:

JSON
1{
2  "agent": {
3    "speak": {
4      "provider": {
5        "type": "eleven_labs",
6        "model_id": "eleven_v3",
7        "language_code": "en-US"
8      },
9      "endpoint": {
10        "url": "https://api.elevenlabs.io/v1/text-to-speech/{voice_id}",
11        "headers": {
12          "xi-api-key": "{{ELEVEN_LABS_API_KEY}}"
13        }
14      }
15    }
16  }
17}

Cartesia

For Cartesia you can use their API to retrieve a voice ID. See their TTS API Docs for more information.

ParameterTypeDescription
agent.speak.provider.typeStringMust be cartesia
agent.speak.provider.model_idStringThe model ID to use
agent.speak.provider.voiceObjectCartesia Voice configuration
agent.speak.provider.voice.modeStringThe voice mode to use
agent.speak.provider.voice.idStringThe voice ID to use
agent.speak.provider.languageStringLanguage setting
agent.speak.endpointObjectMust include url and headers
agent.speak.endpoint.urlStringYour Cartesia API endpoint URL
agent.speak.endpoint.headersObjectHeaders for authentication

Example

JSON
1{
2  "agent": {
3    "speak": {
4      "provider": {
5        "type": "cartesia",
6        "model_id": "sonic-2",
7        "voice": {
8          "mode": "id",
9          "id": "a167e0f3-df7e-4d52-a9c3-f949145efdab"
10        },
11        "language": "en"
12      },
13      "endpoint": {
14        "url": "https://api.cartesia.ai/tts/bytes",
15        "headers": {
16          "x-api-key": "{{CARTESIA_API_KEY}}"
17        }
18      }
19    }
20  }
21}

Amazon (AWS) Polly

For Amazon (AWS) Polly you can refer to this article for a list of available voices.

If no engine is specified, Amazon (AWS) Polly defaults to Standard. If the chosen voice doesn’t support Standard, you’ll get an error like: “Standard engine not supported for {voice}.” In that case, you must explicitly specify the correct engine.

ParameterTypeDescription
agent.speak.provider.typeStringMust be aws_polly
agent.speak.provider.language_codeStringThe language code to use
agent.speak.provider.voiceStringThe voice to use
agent.speak.provider.engineStringThe engine to use
agent.speak.provider.credentialsObjectThe credentials to use

STS Example

JSON
1{
2  "agent": {
3    "speak": {
4      "provider": {
5        "type": "aws_polly",
6        "language_code": "en-US",
7        "voice": "Matthew",
8        "engine": "standard",
9        "credentials": {
10          "type": "sts",
11          "region": "us-west-2",
12          "access_key_id": "{{AWS_ACCESS_KEY_ID}}",
13          "secret_access_key": "{{AWS_SECRET_ACCESS_KEY}}",
14          "session_token": "{{AWS_SESSION_TOKEN}}"
15        }
16      },
17      "endpoint": {
18        "url": "https://polly.us-west-2.amazonaws.com/v1/speech"
19      }
20    }
21  }
22}

IAM Example

JSON
1{
2  "agent": {
3    "speak": {
4      "provider": {
5        "type": "aws_polly",
6        "voice": "Joanna",
7        "language_code": "en-US",
8        "engine": "standard",
9        "credentials": {
10          "type": "iam",
11          "region": "us-east-2",
12          "access_key_id": "{{AWS_ACCESS_KEY_ID}}",
13          "secret_access_key": "{{AWS_SECRET_ACCESS_KEY}}"
14        }
15      },
16      "endpoint": {
17        "url": "https://polly.us-east-2.amazonaws.com/v1/speech"
18      }
19    }
20  }
21}

Using multiple TTS providers

The speak object accepts both a single provider and an array of providers. When you supply an array, the Voice Agent uses the providers as an ordered fallback chain: it sends each TTS request to the first provider in the list and automatically falls back to the next provider if the request fails.

How fallback works

  1. The agent sends the request to the first provider in the array.
  2. If that provider returns an error or times out, the agent sends a SPEAK_REQUEST_FAILED warning over the WebSocket and retries with the next provider.
  3. This continues through every provider in the array.
  4. If all providers fail, the agent sends a FAILED_TO_SPEAK error and the turn produces no audio response.

The fallback is per-request — each new agent utterance starts again from the first provider. Provider order matters, so place your preferred provider first and your most reliable fallback last.

Fallback providers do not need to use the same provider.type. You can mix providers (for example, deepgram primary with an open_ai fallback) to maximize availability across independent infrastructure.

Example

JSON
1{
2  "agent": {
3    "speak": [
4      {
5        "provider": {
6          "type": "deepgram",
7          "model": "aura-2-zeus-en"
8        }
9      },
10      {
11        "provider": {
12          "type": "open_ai",
13          "model": "tts-1",
14          "voice": "shimmer"
15        },
16        "endpoint": {
17          "url": "https://api.openai.com/v1/audio/speech",
18          "headers": {
19            "authorization": "Bearer {{OPENAI_API_KEY}}"
20          }
21        }
22      }
23    ]
24  }
25}

What’s Next