High-Logic/Genie-TTS

GPT-SoVITS ONNX Inference Engine & Model Converter

What it solves

GENIE is a lightweight inference engine designed to make high-quality text-to-speech (TTS) synthesis faster and more accessible, specifically optimizing for CPU performance. It removes the heavy dependencies of the original GPT-SoVITS project, allowing for near-instantaneous speech generation without requiring a powerful GPU.

How it works

Built on the GPT-SoVITS framework, GENIE converts original PyTorch models into an optimized ONNX format. It provides a simplified Python API for loading character voice models, setting reference audio for emotion and intonation cloning, and generating speech. It also includes a built-in FastAPI server for deploying the TTS engine as an API.

Who it’s for

Developers and creators who want to integrate high-quality, multi-language TTS into their applications without the overhead of large model sizes or high latency on CPU-based systems.

Highlights

  • CPU Optimization: Significantly lower first-inference latency and runtime size compared to official PyTorch and ONNX models.
  • Multi-language Support: Supports Japanese, English, Chinese, and Korean.
  • Easy Conversion: Includes a built-in tool to convert .pth and .ckpt GPT-SoVITS models to the optimized ONNX format.
  • Ready-to-use Characters: Provides predefined speaker characters for immediate testing.
  • API Integration: Includes a lightweight FastAPI server for easy deployment.

Related

  • Project
    KittenML/KittenTTS

    A lightweight, ONNX-based text-to-speech library that provides high-quality voice synthesis on CPU with models as small as 25MB.

  • Project
    RVC-Boss/GPT-SoVITS

    A powerful few-shot voice conversion and text-to-speech system that can clone a voice using as little as 5 seconds of audio.

  • Project
    tronghieuit/v-tts

    A lightweight Vietnamese text-to-speech system featuring multi-speaker synthesis and zero-shot voice cloning that runs efficiently on CPU.

  • Project
    OpenMOSS/MOSS-TTS-Nano

    A multilingual tiny speech generation model with 0.1B parameters designed for real-time, CPU-based inference and high-fidelity audio output.

  • Project
    andimarafioti/faster-qwen3-tts

    A high-performance inference engine for Qwen3-TTS that uses CUDA graph capture to enable real-time, low-latency voice cloning and speech generation.