Text-to-speech (TTS) models overview

Overview

Voice agent speech is produced by a TTS model, configured with a voice profile that specifies tone, accent, and other qualitative characteristics of the speech. The TTS model runs on output from an LLM model to speak the agent response to the user.

You can choose a voice model served through LiveKit Inference, included with LiveKit Cloud. With LiveKit Inference, your agent runs on LiveKit's infrastructure to minimize latency. No separate provider API key is required, and usage and rate limits are managed through LiveKit Cloud. Use the plugin instead if you prefer to manage billing and rate limits yourself, or need access to a provider not currently available through LiveKit Inference.

LiveKit Inference

LiveKit Inference provides a curated set of TTS models with managed billing and rate limits. Each model comes with a limited selection of Suggested voices, plus a wider selection through each provider's documentation.

Models

The following models are available in LiveKit Inference. Refer to the guide for each model for more details on additional configuration options.

Custom voices

You can create and use custom voices with LiveKit Inference. Upload or record a sample, and LiveKit clones it to all supported TTS providers on your plan. You can then use the clone in your agent sessions with any of those providers.

Suggested voices

The following voices are good choices for overall quality and performance. Each provider has a much larger selection of voices to choose from, which you can find in their documentation. In addition to the voices below, you can choose to use other TTS provider voices through LiveKit Inference.

Click the copy icon to copy the voice ID to use in your agent session.

Plugins

The LiveKit Agents framework includes open source plugins for a wide range of TTS providers. Use a plugin when you need provider-specific features not available through Inference, want to manage billing directly, or need a provider not currently in Inference. Plugins require your own API key and account.

Have another provider in mind? LiveKit is open source and welcomes new plugin contributions.

TTS Usage

To set up TTS in an AgentSession, provide a descriptor with both the desired model and voice. LiveKit Inference manages the connection to the model automatically. Consult the Suggested voices list for suggested voices, or view the model reference for more voices.

from livekit.agents import AgentSession, inference

session = AgentSession(
    tts=inference.TTS(
        model="cartesia/sonic-3",
        voice="9626c31c-bec5-4cca-baa8-f8ba9e84c8bc",
        language="en",
    ),
    # ... llm, stt, etc.
)

import { AgentSession, inference } from '@livekit/agents';

const session = new AgentSession({
    tts: new inference.TTS({
        model: "cartesia/sonic-3",
        voice: "9626c31c-bec5-4cca-baa8-f8ba9e84c8bc",
        language: "en",
    }),
    // ... llm, stt, etc.
})

Additional parameters

More configuration options, such as custom pronunciation, are available for each model. To set additional parameters, use the TTS class from the inference module. Consult each model reference for examples and available parameters.

Language codes

All TTS plugins and LiveKit Inference use the LanguageCode type for the language parameter. LanguageCode accepts any common language format and normalizes it automatically to BCP-47 . You don't need to look up the specific format each provider expects — pass any of the following and the framework handles the conversion:

• ISO 639-1: "en", "es", "fr"
• BCP-47 with region: "en-US", "zh-Hans-CN"
• ISO 639-3: "eng", "spa"
• Language names: "english", "spanish"
• Underscored variants: "en_us" (normalized to "en-US")

For example, all of the following are equivalent:

from livekit.agents import LanguageCode

LanguageCode("english")  # → "en"
LanguageCode("eng")      # → "en"
LanguageCode("en")       # → "en"
LanguageCode("en-US")    # → "en-US"
LanguageCode("en_us")    # → "en-US"

import { normalizeLanguage } from '@livekit/agents';

normalizeLanguage("english")  // → "en"
normalizeLanguage("eng")      // → "en"
normalizeLanguage("en")       // → "en"
normalizeLanguage("en-US")    // → "en-US"
normalizeLanguage("en_us")    // → "en-US"

Custom TTS

To create an entirely custom TTS, implement the TTS node in your agent.

Standalone TTS

You can use a TTS instance as a standalone component by creating a stream. Use push_text to add text to the stream, and then consume a stream of SynthesizedAudio to publish as realtime audio to another participant.

Here is an example of a standalone TTS app:

import asyncio
from livekit import agents, rtc
from livekit.agents import AgentServer
from livekit.agents.tts import SynthesizedAudio
from livekit.plugins import cartesia
from typing import AsyncIterable


server = AgentServer()

@server.rtc_session(agent_name="my-agent")
async def my_agent(ctx: agents.JobContext):
    text_stream: AsyncIterable[str] = ... # you need to provide a stream of text
    audio_source = rtc.AudioSource(44100, 1)

    track = rtc.LocalAudioTrack.create_audio_track("agent-audio", audio_source)
    await ctx.room.local_participant.publish_track(track)

    tts = cartesia.TTS(model="sonic-english")
    tts_stream = tts.stream()

    # create a task to consume and publish audio frames
    asyncio.create_task(send_audio(tts_stream))

    # push text into the stream, TTS stream will emit audio frames along with events
    # indicating sentence (or segment) boundaries.
    async for text in text_stream:
        tts_stream.push_text(text)
    tts_stream.end_input()

    async def send_audio(audio_stream: AsyncIterable[SynthesizedAudio]):
        async for a in audio_stream:
            await audio_source.capture_frame(a.audio.frame)

if __name__ == "__main__":
    agents.cli.run_app(server)

Additional resources

The following resources cover related topics that may be useful for your application.