Soniox TTS Now Available for Telnyx Voice AI

Soniox text-to-speech is now available for Telnyx Voice AI, adding about 200 voices that each speak 60 languages from one model to the voice provider catalog.

What's new

• Soniox TTS provider: Select Soniox voices using the Soniox.tts-rt-v2.<Name> format, with about 200 voices in the catalog including Soniox.tts-rt-v2.Adrian, Soniox.tts-rt-v2.Emma, and Soniox.tts-rt-v2.Nina.
• 60 languages from one model: Every Soniox voice speaks every supported language, including English, Spanish, French, German, Portuguese, Hindi, Japanese, and Arabic. Set language to the two-letter ISO 639-1 code of the text, with en as the default. On Voice AI Assistants, one voice follows the caller when they change language during a call.
• Precise alphanumeric rendering: Soniox built the model to render phone numbers, email addresses, IDs, verification codes, and account numbers accurately.
• Low-latency streaming synthesis: Stream speech over WebSocket or REST with telephony-native 8 kHz pcm_mulaw and pcm_alaw output.
• Available across Telnyx voice surfaces: Use Soniox voices in Voice AI Assistants, the Telnyx TTS API (POST /v2/text-to-speech/speech and WebSocket streaming), Call Control speak, gather_using_speak, and conference speak commands, and TeXML <Say>.
• Speed and pacing controls: Set speed from 0.7 to 1.3, and set reduce_silence to shorten the pauses between words.
• Plain text input: Write prompts as plain text. SSML is not supported, and each request is capped at 2 minutes of audio.

Why it matters

• Telnyx is the infrastructure voice agents run on: telephony, STT, TTS, and agent orchestration in one platform, instead of a Frankenstack of four to six vendors stitched together.
• Soniox STT is already available on Telnyx, so teams can now run both sides of the conversation, listening and speaking, with one speech AI provider.
• Every voice speaks every supported language, so teams build one agent voice that serves callers across 60 languages with no per-language model switching or voice swapping.
• Teams serving callers in several markets can give an agent one consistent voice across all of them, instead of picking a different voice or vendor per language.

Example use cases

• Support agents that read verification codes, account numbers, and contact details back to callers with accurate alphanumeric rendering.
• Global contact centers serving callers in 60 languages from a single voice, including mixed-language conversations.
• Developers building conversational agents that pair Soniox STT with Soniox TTS for consistent speech quality across listening and speaking.
• Product teams adding voices to Call Control or TeXML call flows without adding another vendor to the stack.

Getting started

1. In Mission Control, navigate to AI, then Assistants, then select your assistant.
2. Open the Voice tab.
3. Choose Soniox as the provider and pick a voice.
4. Save the assistant and test with real caller audio before routing production traffic.

For API-based call flows, pass a Soniox voice to POST /v2/text-to-speech/speech, over WebSocket streaming, or in a Call Control or TeXML call. Set language to the language of the text. Soniox lists the full set of supported languages. Example REST request:

{
  "text": "Your verification code is 4729. We sent it to [email protected].",
  "voice": "Soniox.tts-rt-v2.Adrian",
  "voice_settings": {
    "language": "en"
  }
}

In TeXML, set the voice and language on <Say>:

<Response>
  <Say voice="Soniox.tts-rt-v2.Emma" language="it">Buongiorno, come posso aiutarla?</Say>
</Response>

Learn more in the Soniox provider docs, the Available Voices catalog, or the Telnyx Text-to-Speech overview.