livekit-plugins-respeecher 0.1.0

Respeecher TTS plugin for LiveKit Agents.

Respeecher LiveKit Plugin

Respeecher TTS integration for LiveKit Agents.

This package is maintained by Respeecher. It uses the normal LiveKit plugin import path, but it is distributed outside the LiveKit monorepo while the upstream PR is under review.

Installation

pip install livekit-agents livekit-plugins-respeecher

Configure

export LIVEKIT_URL="wss://..."
export LIVEKIT_API_KEY="..."
export LIVEKIT_API_SECRET="..."
export RESPEECHER_API_KEY="..."

RESPEECHER_API_KEY is used when api_key is not passed to the constructor; you can also pass it directly: respeecher.TTS(api_key="...", voice_id="...").

Usage

from dotenv import load_dotenv

from livekit.agents import Agent, AgentSession, JobContext, WorkerOptions, cli
from livekit.plugins import respeecher

load_dotenv()


async def entrypoint(ctx: JobContext) -> None:
    await ctx.connect()

    session = AgentSession(
        tts=respeecher.TTS(voice_id="marta"),
    )

    await session.start(
        room=ctx.room,
        agent=Agent(instructions="You are a concise voice assistant."),
    )


if __name__ == "__main__":
    cli.run_app(WorkerOptions(entrypoint_fnc=entrypoint))

voice_id is required because each Respeecher model exposes its own voice list. The plugin streams over WebSocket by default; chunked HTTP synthesis is also available via TTS.synthesize(text).

Discover Voices

import asyncio

from livekit.plugins import respeecher


async def main() -> None:
    tts = respeecher.TTS(voice_id="placeholder")
    voices = await tts.list_voices()
    for voice in voices:
        print(voice.id)


asyncio.run(main())

Use a voice_id returned by list_voices() for the model you plan to use.

Overriding Sampling Parameters

See the Sampling Parameters Guide.

from livekit.plugins import respeecher

tts = respeecher.TTS(
    voice_id="marta",
    voice_settings=respeecher.VoiceSettings(
        sampling_params={
            "min_p": 0.01,
        },
    ),
)

Models

See Models & Languages. Supported models:

• /public/tts/en-rt (default)
• /public/tts/ua-rt

To use the Ukrainian model, pass it explicitly:

from livekit.plugins import respeecher

tts = respeecher.TTS(
    voice_id="olesia-conversation",
    model="/public/tts/ua-rt",
)

Compatibility

Requires livekit-agents>=1.5,<2.

Support

For plugin issues, open an issue in this repository; for LiveKit Agents itself, use the LiveKit repo. If this plugin helps your use case, feedback on the upstream PR is appreciated.