Two ways to run it
Use LiveKit Inference to run Fish Audio on LiveKit's infrastructure with no separate provider API key, or install the plugin to manage your own Fish Audio billing and rate limits.
Real-time voice agents with Fish Audio on LiveKit
Fish Audio text-to-speech ships as a first-party plugin for LiveKit Agents and through LiveKit Inference, so your agents speak with Fish Audio's most expressive voices at conversational latency.
CategoryReal-time infrastructure
IntegrationLiveKit Agents plugin (Python) and LiveKit Inference
StreamingChunked and real-time WebSocket
Default modelFish Audio S2.1 Pro
VoicesAny Fish Audio voice or clone by reference_id
docs.livekit.io ↗Why LiveKit and Fish Audio
LiveKit handles the real-time transport, turn detection and agent orchestration. Fish Audio handles how the agent sounds. Together they are one of the fastest paths from prototype to a production voice agent.
Streaming built for turn-taking
The plugin supports chunked synthesis and real-time WebSocket streaming, so speech starts as soon as the first tokens arrive instead of waiting for the full response.
Every Fish Audio voice, tunable per agent
Point reference_id at any voice in the Fish Audio library or a clone you made, then set speed, volume, temperature and top_p per agent.
How it works
3 steps from a Fish Audio voice to a shipping product on LiveKit.
STEP 01
Install the plugin
Add Fish Audio to LiveKit Agents with a single extra.
pip install "livekit-agents[fishaudio]"STEP 02
Get API KeySet your API Key
Export FISH_API_KEY in the agent's environment, or skip this step entirely with LiveKit Inference.
STEP 03
Add TTS to your agent
Import the TTS class, pick a voice and model, and hand it to your AgentSession. Real-time streaming is one flag away.
agent.py
1234567
from livekit.plugins.fishaudio import TTS
tts = TTS(
reference_id="your_voice_model_id",
model="s2.1-pro",
sample_rate=24000,
latency_mode="balanced",
)What teams build with Fish Audio and LiveKit
Customer support agents
Voice agents that answer, triage and resolve in real time, with a consistent brand voice across every call.
Companions and characters
Expressive, emotionally controllable voices for games, companions and interactive characters.
Multilingual assistants
One agent, many markets. Fish Audio voices cover the major world languages inside a single LiveKit pipeline.
More partners
AI video platform. Fish Audio powers the Fish voice engine inside HeyGen's AI Studio, with accent and locale controls for English. Available in both HeyGen Video Creator and LiveAvatar.
Video & avatars
Voice agent platform for phone and call centers. Import any Fish Audio voice to power your Retell agent.
Voice agent platforms
Global execution layer for real-time voice AI. Runs Fish Audio S2.1 Pro on in-region GPUs in Europe and Australia.
Regional inference
Start building with Fish Audio and LiveKit
Read the LiveKit integration guide, or talk to our partnerships team about co-marketing and joint customers.