For AI agents: a documentation index is available at /llms.txt. Markdown versions of all pages can be requested by appending `.md` to the URL, or by setting the `Accept` header to `text/markdown`.
Skip to main content
Speech to TextAgent STT

Overview

Turn-based transcription for building voice agents.

Agent STT is transcription designed for building voice agents. Stream audio in and receive speaker-labelled, turn-based transcription back — punctuated and ready to pass to an LLM.

Agent STT is available on SaaS only.

There are three ways to use it:

Integrations

The fastest path to a production voice agent, on LiveKit, Pipecat or Vapi

Agent STT SDK

A Python client that handles the connection, audio streaming and message handling

Agent STT API

Connect over WebSocket from any language, for custom pipelines or unsupported platforms

Features​

  • Turn detection: detect when a speaker has finished talking. See Turn detection.
  • Intelligent segmentation: group words into clean, speaker-attributed segments. See Segmentation.
  • Diarization: identify and label different speakers.
  • Speech signals: receive signals for start and end of speech.

Integrations​

Use an integration to handle audio transport and wiring, so you can focus on your agent logic:

Vapi logo

Vapi

Turnkey voice agent platform. Deploy fast with no code.
LiveKit logo

LiveKit

Open-source framework for building agents with WebRTC infrastructure.
Pipecat logo

Pipecat

Open-source framework with full control of the voice pipeline in code.

Agent STT SDK​

The Agent STT SDK is a Python client for the Agent STT service. It manages the WebSocket session, streams your audio, and delivers segments and turn events to your callbacks.

See the Quickstart to get started.

Using the API directly​

Connect to the WebSocket directly from any language when there's no SDK for your stack, or when you need control over the session that a client library doesn't expose.

wss://global.rt.speechmatics.com/v2/agent

Use the global endpoint to route to the nearest region, or pin to a region for data residency. See Supported endpoints for regional endpoints.

Open the connection with your API key, send StartRecognition with your audio format and transcription config, then stream binary audio in AddAudio messages. Segments and turn events arrive as JSON while you send.

See the Agent STT Reference for the session flow, every message schema and the error codes.