Rime TTS | LiveKit Documentation

Create a new agent in your browser using this model

Overview

Rime text-to-speech is available in LiveKit Agents through LiveKit Inference and the Rime plugin. 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 want to manage your own billing and rate limits. Pricing for LiveKit Inference is available on the pricing page .

LiveKit Inference

Use LiveKit Inference to access Rime TTS without a separate Rime API key.

Model name	Model ID	Languages
Arcana	rime/arcana	enesfrdehihejaptar
Coda	rime/coda	enesfrdeptja
Mist	rime/mist	en
Mist v2	rime/mistv2	enesfrde
Mist v3	rime/mistv3	enesfrdehi

Usage

To use Rime, use the TTS class from the inference module:

from livekit.agents import AgentSession, inference

session = AgentSession(
    tts=inference.TTS(
        model="rime/arcana",
        voice="celeste",
        language="en"
    ),
    # ... tts, stt, vad, turn_handling, etc.
)

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

session = new AgentSession({
    tts: new inference.TTS({
        model: "rime/arcana",
        voice: "celeste",
        language: "en"
    }),
    // ... tts, stt, vad, turnHandling, etc.
});

Parameters

model

Required

string

The model ID from the models list.

voice

Required

string

See voices for guidance on selecting a voice.

languageLanguageCode

Language code for the input text. The framework automatically converts to the format expected by the Rime API.

extra_kwargsdict

Additional parameters to pass to the Rime TTS API. See model parameters for supported fields.

In Node.js this parameter is called modelOptions.

Model parameters

Pass the following parameters inside extra_kwargs (Python) or modelOptions (Node.js). Each parameter applies only to the models listed in the Models column.

Parameter	Type	Default	Models	Notes
`speed_alpha`	`float`	`1.0`	`mist`, `mistv2`, `mistv3`	Speed multiplier. Values less than `1.0` produce faster speech; values greater than `1.0` produce slower speech.
`pause_between_brackets`	`bool`	`False`	`mist`, `mistv2`, `mistv3`	Whether to insert a pause at bracket positions in the text.
`phonemize_between_brackets`	`bool`	`False`	`mist`, `mistv2`, `mistv3`	Whether to interpret text between brackets as phonemes for custom pronunciation .
`inline_speed_alpha`	`str`		`mist`, `mistv2`, `mistv3`	Comma-separated list of speed multipliers applied inline to segments of text.
`no_text_normalization`	`bool`	`False`	`mist`, `mistv2`, `mistv3`	Whether to skip text normalization (disabling expansion of numbers, abbreviations, etc.). Might reduce latency but can affect pronunciation of digits and abbreviations.
`time_scale_factor`	`float`	`1.0`	`arcana`, `coda`, `mistv3`	Controls audio playback speed. Values greater than `1.0` slow the audio; values less than `1.0` speed it up. Serialized as `timeScaleFactor` to match the Rime API.

Voices

LiveKit Inference supports all of the voices available in the Rime API. You can view the default voices and explore the wider set in the API in the Rime voices documentation , and use the voice by copying its name into your LiveKit agent session.

The following is a small sample of the Rime voices available in LiveKit Inference.

Use the keyboard and arrows to audition voices

String descriptors

As a shortcut, you can also pass a descriptor with the model ID and voice directly to the tts argument in your AgentSession:

from livekit.agents import AgentSession

session = AgentSession(
    tts="rime/arcana:celeste",
    # ... llm, stt, vad, turn_handling, etc.
)

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

session = new AgentSession({
    tts: "rime/arcana:celeste",
    // ... tts, stt, vad, turnHandling, etc.
});

Plugin

LiveKit's plugin support for Rime lets you connect directly to Rime's TTS API with your own API key.

Available inPython

Node.js

Installation

Install the plugin:

uv add "livekit-agents[rime]~=1.5"

pnpm add @livekit/agents-plugin-rime@1.x

Authentication

The Rime plugin requires a Rime API key .

Set RIME_API_KEY in your .env file.

Usage

Use Rime TTS within an AgentSession or as a standalone speech generator. For example, you can use this TTS in the Voice AI quickstart.

from livekit.plugins import rime

session = AgentSession(
   tts=rime.TTS(
      model="arcana",
      speaker="celeste",
      speed_alpha=0.9,
   ),
   # ... llm, stt, etc.
)

import * as rime from '@livekit/agents-plugin-rime';

const session = new voice.AgentSession({
    tts: new rime.TTS({
      modelId: "arcana",
      speaker: "celeste",
      speedAlpha: 0.9,
    }),
    // ... llm, tts, etc.
});

WebSocket streaming

By default, the plugin synthesizes speech over Rime's HTTP API. Set use_websocket=True to opt into WebSocket streaming, which lowers latency and emits word-level timestamps for TTS-aligned transcriptions. Streaming and aligned transcripts are enabled automatically when WebSocket mode is active.

from livekit.plugins import rime

session = AgentSession(
   tts=rime.TTS(
      model="arcana",
      speaker="celeste",
      use_websocket=True,
      segment="bySentence",
   ),
   # ... llm, stt, etc.
)

import * as rime from '@livekit/agents-plugin-rime';

const session = new voice.AgentSession({
    tts: new rime.TTS({
      modelId: "arcana",
      speaker: "celeste",
      useWebsocket: true,
      segment: "bySentence",
    }),
    // ... llm, tts, etc.
});

Parameters

This section describes some of the available parameters. See the plugin reference for a complete list of all available parameters.

Coda model

When model="coda", the default speaker is lyra. The coda model ignores Rime parameters such as speed_alpha, reduce_latency, pause_between_brackets, phonemize_between_brackets, temperature, top_p, and repetition_penalty.

modelstringDefault: arcana

ID of the model to use. To learn more, see Models .

speakerstring

ID of the voice to use for speech generation. If unset, the SDK selects a default voice based on the model. To learn more, see Voices .

audio_formatTTSEncodingDefault: pcm

Audio format to use. Valid values are: pcm and mp3.

sample_rateintegerDefault: 16000

Sample rate of the generated audio. Set this rate to best match your application needs. To learn more, see Recommendations for reducing response time .

speed_alphafloatDefault: 1.0

Adjusts the speed of speech. Lower than 1.0 results in faster speech while higher than 1.0 results in slower speech. Supported on the mist models (mist, mistv2, mistv3) only.

time_scale_factorfloatDefault: 1.0

Controls audio playback speed. Values greater than 1.0 slow the audio while values less than 1.0 speed it up. Supported on the arcana, coda, and mistv3 models only. Passing it with mistv2 raises an error.

In Node.js, this parameter is called timeScaleFactor.

reduce_latencybooleanDefault: false

When set to true, turns off text normalization to reduce the amount of time spent preparing input text for TTS inference. This might result in the mispronunciation of digits and abbreviations. Supported on mistv2 only. To learn more, see Recommendations for reducing response time .

phonemize_between_bracketsbooleanDefault: false

When set to true, allows the use of custom pronunciation strings in text. To learn more, see Custom pronunciation . Supported on the mist models (mist, mistv2, mistv3) only.

api_keystringEnv: RIME_API_KEY

Rime API Key. Required if the environment variable isn't set.

use_websocketbooleanDefault: false

Opts into WebSocket streaming. When false, the plugin uses Rime's HTTP API. WebSocket mode enables streaming and word-level timestamps automatically. To learn more, see WebSockets .

In Node.js, this parameter is called useWebsocket.

base_urlstring

Overrides the Rime endpoint URL. Defaults to https://users.rime.ai/v1/rime-tts for HTTP synthesis and wss://users-ws.rime.ai for WebSocket streaming. Passing a ws:// or wss:// URL sets use_websocket=True.

In Node.js, this parameter is called baseURL.

segmentstringDefault: bySentence

Requires use_websocket=True. Controls server-side text segmentation:

"bySentence" (default) synthesizes at sentence boundaries.
"immediate" synthesizes as soon as text arrives in the buffer.
"never" waits for explicit flush commands.

To learn more, see Segment values .

tokenizerSentenceTokenizer

Requires use_websocket=True. Client-side sentence tokenizer used to split text into sentences before streaming to Rime. Defaults to the SDK's built-in sentence tokenizer.

Customizing pronunciation

Rime TTS supports customizing pronunciation. To learn more, see Custom Pronunciation guide .

Additional resources

The following resources provide more information about using Rime with LiveKit Agents.

Python

PyPI Reference GitHub

Node.js

NPM Reference GitHub

Rime docs

Rime TTS docs.

Voice AI quickstart

Get started with LiveKit Agents and Rime TTS.