Kokoros provides an extremely fast, high-quality Text-to-Speech (TTS) inference engine implemented in Rust, based on the popular Kokoro model. It targets developers and users seeking real-time, embeddable TTS capabilities, offering significant performance gains over Python-based implementations.

How It Works

Kokoros leverages Rust for its performance and memory safety, enabling efficient inference of the 87M parameter Kokoro model. It integrates a phonemizer, removing external dependencies for end-to-end synthesis. The project supports multiple languages (English, Chinese, Japanese, German) with ongoing expansion plans.

Quick Start & Requirements

• Install Python dependencies: pip install -r scripts/requirements.txt
• Fetch voice data: python scripts/fetch_voices.py
• Build the Rust project: cargo build --release
• Usage: ./target/release/koko [options]
• Docker image available for easier deployment.
• See official documentation for detailed usage and server setup.

Highlighted Details

• Streaming mode supported for real-time audio output.
• Style mixing for ASMR effects and voice variation.
• OpenAI-compatible API server for seamless integration.
• Espeak-ng tokenizer and phonemizer support for end-to-end synthesis.

Maintenance & Community

• Active development with recent updates including streaming, style mixing, and OpenAI server support.
• Discord community available for discussion and support: https://discord.gg/E566zfDWqD

Licensing & Compatibility

• Licensed under the Apache License.
• Compatible with commercial use and closed-source linking.

Limitations & Caveats

The project is under active development, with some language support noted as "partly" implemented. The OpenAI server compatibility is still undergoing polish.