Efficient AI text-to-speech model with 82M parameters delivering natural-sounding voice synthesis
Photorealistic avatars that speak your voice agent's audio in real time
Stage
Not given
Not given
Founded
Not given
Not given
Based in
SG
Not given
Pricing model
Open source
Paid
Pricing
Free to use. Free Chrome Extension with 10,000 characters free every month.
Realistic Avatars: $0.02 per minute, billed per second. Voice Agents: $0.06 per minute, billed per second. Open Source: Free (MIT License). Custom avatar: $199 per avatar (one time). Remove watermark: $19 per month. New accounts get $1 of free credits.
Key features
82M parameter efficiency for lightweight, scalable performance
Multilingual support including English, French, Korean, Japanese, and Mandarin
Customizable voice packs with multiple lifelike voice options
Automatic content segmentation for chapter and section detection
OpenAI-compatible API endpoint for easy integration
Real-time audio generation with NVIDIA GPU acceleration
Processes up to 510 tokens in a single pass for longer audio outputs
Chrome extension for reading webpages aloud
Real-time lip-sync with accurate mouth shapes from audio alone
Works with any TTS or voice provider (OpenAI, ElevenLabs, Azure, Fish Audio, Cartesia)
Avatar responds in less than 250 milliseconds
WebRTC delivery with video and audio in sync, no media servers required
LiveKit integration for drop-in avatar as participant
SDK with server and client libraries for quick integration
User interruption support to stop avatar mid-sentence
Sessions can run up to 3 hours each
Works with any language audio
No text or viseme data required
What makes it different
Outperforms larger models like XTTS (467M params) and MetaVoice (1.2B params) despite having only 82M parameters
Open-source and free for both commercial and personal use under Apache 2.0 license
Lightweight architecture enabling deployment on CPU and GPU with minimal resources
Trained on high-quality, permissively licensed audio for accurate natural-sounding speech
Designed for scalability while preserving excellent audio quality
Photorealistic avatars that require no 3D work or custom training
Ultra-low latency under 250 milliseconds
Affordable pricing from $0.02 per minute, billed per second with no minimum
Kokoro TTS
Talking Avatar