Speechify plugin for LiveKit Agents

The Agents Framework is designed for building realtime, programmable participants that run on servers. Use it to create conversational, multi-modal voice agents that can see, hear, and understand.

This package contains the Speechify plugin, which provides streaming text-to-speech with word-level timestamps. Refer to the documentation for information on how to use it. See the repository for more information about the framework as a whole.

Installation

pnpm add @livekit/agents-plugin-speechify

Usage

Set SPEECHIFY_API_KEY in your environment, or pass apiKey. Get a key at platform.speechify.ai.

import { voice } from '@livekit/agents';
import * as speechify from '@livekit/agents-plugin-speechify';

const session = new voice.AgentSession({
  tts: new speechify.TTS({ voiceId: 'dominic_32', model: 'simba-3.2' }),
  // ... stt, llm, etc.
});

simba-3.2 is English only and has the lowest time to first audio; use simba-3.0 for the other languages it supports. The voice must support the model; list voices with GET /v1/voices.

stream() splits its input into sentences and synthesizes each one as soon as it is complete, over POST /v1/audio/stream/with-timestamps. Audio (24 kHz mono PCM) and word timestamps are forwarded as they arrive, so captions stay aligned with speech. Requests share keep-alive connections, which prewarm() opens before the first reply.