---
title: "Conversation Relay: Plug voice into your text AI workflow | Telnyx"
canonical: https://telnyx.com/products/conversation-relay
---
# Conversation Relay: Plug voice into your text AI workflow | Telnyx

**VOICE API**

# Conversation Relay

You handle the text, we handle the media. Conversation Relay streams transcribed voice as text over WebSockets, so you can plug live voice into the AI engine you already run, no audio handling required. Telnyx manages transport, STT, and TTS.

[Start building](https://telnyx.com/sign-up)[Contact us](https://telnyx.com/contact-us)

WHY TELNYX

## Best-of-breed speech, on a network built for voice

You bring the intelligence. Telnyx brings the speech layer and the line underneath it.

* ### Best-of-breed STT and TTS  
Choose from a range of speech-to-text engines and a catalog of voices. Pick the voice and the accuracy your use case needs, not whatever one vendor ships.
* ### Low-latency media  
STT and TTS run close to the call, so the audio layer stays fast and the conversation feels natural. Your engine's response time is yours; the media around it is ours.
* ### Global reach  
Full programmatic control over PSTN voice services in 140+ countries. Your voice runs on Telnyx's own carrier network, not a resold one. As a licensed carrier, we deliver secure, compliant, and reliable infrastructure.

DEMO

## Configure the environment your agents run in

Try different STT, TTS engines and connect a phone number to your agent

Global NumbersText to SpeechSpeech to Text

Loading...

**HOW IT WORKS**

## You own the logic. We own the line.

Telnyx transcribes the caller, streams you the text, and speaks your reply back. Your side of the connection is text from start to finish.

THE DIVISION OF LABOR

You keep full control of how your agent thinks and responds. Everything from the phone number to the spoken word is on us. Here is the exact split.

In your stack

### You handle

* Your LLM and reasoning
* Your knowledge base
* Your tools and function calling
* Your conversation logic

On our network

### Telnyx handles

* Phone numbers and carrier network
* Voice transport
* Real-time speech-to-text
* Text-to-speech and voices
* DTMF, interruption, and call control

## One WebSocket between Telnyx and your application

A single bidirectional connection per session carries the whole exchange as text frames. Open it from TeXML or the Voice API, point it at your endpoint, and you are live.

[Explore dev docs](https://developers.telnyx.com/docs/voice/programmable-voice/conversation-relay)

TeXML

```
<?xml version="1.0" encoding="UTF-8"?>
<Response>
  <Connect>
    <ConversationRelay
      url="wss://yourdomain.com/conversation-relay"
      voice="Telnyx.Ultra.01eaafa9-308a-4276-a017-6ab0cf061b1f"
      language="en"
      transcriptionProvider="deepgram"
      welcomeGreeting="Welcome! How can I help you today?"
    />
  </Connect>
</Response>
```

Voice API

```
curl -X POST https://api.telnyx.com/v2/calls/{call_control_id}/actions/conversation_relay_start \
  --header "Content-Type: application/json" \
  --header "Authorization: Bearer ***" \
  --data '{
    "url": "wss://yourdomain.com/conversation-relay",
    "voice": "Telnyx.Ultra.01eaafa9-308a-4276-a017-6ab0cf061b1f",
    "language": "en-US",
    "transcription_engine": "Deepgram",
    "greeting": "Welcome! How can I help you today?"
  }'
```

TeXML multi-language

```
<?xml version="1.0" encoding="UTF-8"?>
<Response>
  <Connect>
    <ConversationRelay
      url="wss://yourdomain.com/conversation-relay"
      voice="Telnyx.Ultra.01eaafa9-308a-4276-a017-6ab0cf061b1f"
      language="en"
      transcriptionProvider="deepgram"
      welcomeGreeting="Press 1 for English, 2 for French, 3 for Spanish."
      dtmfDetection="true"
    >
      <Language code="fr" voice="Telnyx.Ultra.0d09e991-5763-406e-b637-02bc431ef72d" transcriptionProvider="google" />
      <Language code="es" voice="Telnyx.Ultra.13ff5deb-2591-42ad-a356-63a04e524411" transcriptionProvider="telnyx" />
    </ConversationRelay>
  </Connect>
</Response>
```

WebSocket

```
// Telnyx sends the caller's speech as text
{ "type": "prompt", "voicePrompt": "what are your hours", "lang": "en", "last": true }

// Your app sends text back to speak
{ "type": "text", "token": "We're …day.", "last": true }
```

TeXMLVoice APITeXML multi-languageWebSocket

**PRICING**

## Transparent pricing, pay for what you use

Starting at $0.05 per minute. You bring your own AI engine, so there is no model or platform fee from Telnyx for the reasoning layer.

$0.05

Starting cost per minute

[See pricing](https://telnyx.com/pricing/voice-api)

PRODUCTS

## The rest of the stack, when you want it

Conversation Relay is one way in. When you are ready for Telnyx to run the model too, the Voice AI Platform, Inference, Speech to Text, and Text to Speech all sit on the same network, same API key, same bill.

[See all products](https://telnyx.com/products)

* ### Text to Speech  
One API for leading voice engines. Access a catalog of natural and HD voices without lock-in.  
[Learn more](https://telnyx.com/products/text-to-speech-api)
* ### Speech to Text  
Real-time transcription with best-of-breed engines. Choose accuracy, speed, and vendor for your use case.  
[Learn more](https://telnyx.com/products/speech-to-text-api)
* ### Voice AI Platform  
Build and deploy low-latency Voice AI agents in minutes on a full-stack, conversational AI platform.  
[Learn more](https://telnyx.com/products/voice-ai-agents)
* ### Inference  
Serverless inference on GPUs Telnyx owns. Run open-source and frontier models on the same network, API key, and bill.  
[Learn more](https://telnyx.com/products/inference)
* ### Deepfake Detection  
Detect AI-generated voice fraud in real time, built into the Telnyx voice network.  
[Learn more](https://telnyx.com/products/deepfake-detection)

## Sign up and start building

Keep your AI engine. Add voice without the audio pipeline.

[Start building](https://telnyx.com/sign-up)[Contact us](https://telnyx.com/contact-us)

## FAQ

### What is Conversation Relay?

Conversation Relay connects a live Telnyx call to your WebSocket application. Telnyx handles speech recognition and text-to-speech; your application receives the caller's words as text and sends text back to be spoken. It lets you add voice to a text-based AI workflow without processing raw audio.

### How is it different from media streaming?

### Can I use my own AI engine?

### What does Telnyx handle, and what does my application handle?

### What workflows is it best suited for?

### How is this different from the Telnyx Voice AI Platform?

### What voices and languages are supported?

[Start building](https://telnyx.com/sign-up)