KittenML/KittenTTS

Open-source State-of-the-art TTS model which runs on a CPU 😻

What it solves

Kitten TTS is an open-source text-to-speech (TTS) library that provides high-quality, natural-sounding audio generation from text. It addresses the need for efficient, expressive speech synthesis that can run on consumer-grade hardware, including CPUs, without requiring a GPU.

How it works

The library features two main model families: KittenTTS 2, a 1.7B-parameter 1-bit speech language model, and a series of lightweight ONNX models (15M-80M parameters). KittenTTS 2 enables in-context voice cloning by using a 5-30 second audio clip of a target speaker as a reference. It uses a dual-route approach for identity preservation, combining a speaker embedding and codec tokens in the prompt. The system also supports multilingual synthesis across 20 languages and includes text preprocessing to automatically expand numbers, dates, and abbreviations.

Who it’s for

This tool is designed for developers and creators who need to integrate speech synthesis into their applications, particularly those who need voice cloning and emotional expression control without the need for expensive GPU infrastructure.

Highlights

  • Zero-shot voice cloning: Clone any voice from a short audio sample (5-30 seconds) without fine-tuning.
  • CPU-optimized: Optimized C++ inference via a llama.cpp fork for real-time performance on CPUs.
  • Expression control: Use tags for emotions, vocal events (like laughing), and emphasis to steer delivery.
  • Multilingual support: Supports 20 different languages.
  • Long-form text handling: Automatic sentence-aware chunking and seamless joining for long scripts.
  • Streaming: Supports chunk-level streaming to reduce initial playback delay.
  • Extensive built-in voices: Includes 47 pre-defined voices.

Related

  • Project
    thewh1teagle/kokoro-onnx

    A lightweight ONNX Runtime implementation of Kokoro-TTS that enables fast, multi-language text-to-speech generation on CPU and GPU.

  • 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 voice cloning.

  • 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
    kyutai-labs/pocket-tts

    Pocket TTS is a 100 M‑parameter, CPU‑only text‑to‑speech library that streams audio with ~200 ms latency, supports multiple languages and voice cloning, and offers a Python API, CLI, and local HTTP server. It can be installed via pip (or uv) and runs faster than real‑time on modest hardware, with community‑trained models and several alternative runtimes for browsers and embedded devices.

  • Project
    HojoAI/Hojo-TTS-Light

    A lightweight, bilingual (Chinese and English) text-to-speech model that enables high-quality speech synthesis and voice cloning on CPUs and edge devices.