Kitten TTS 😻 Rust fork 🦀

Kitten TTS is an open-source realistic text-to-speech model with just 15 million parameters, designed for lightweight deployment and high-quality voice synthesis.

Currently in developer preview (Original author note)

Checkout original project git

✨ Features

• Ultra-lightweight: Model size less than 25MB
• CPU-optimized: Runs without GPU on any device
• High-quality voices: Several premium voice options available
• Fast inference: Optimized for real-time speech synthesis

🦀 Fork Features

• CLI
• No hard espeak dependency
• Option of using any phonemizer

🚀 Quick Start

Usage CLI

# Build package using Nix package manager
nix build

Usage Lib

# Create development environment using Nix package manager
nix develop

Basic Lib Usage

let model = crate::KittenModel::model_builtin(crate::KittenVoice::default());
let inference = model
    .unwrap()
    .generate("This high quality TTS model works without a GPU".to_string());
let (waveform, _) = inference.unwrap();
let wav = crate::wav::save_array1_f32_as_wav(&waveform, "out/out.wav", None);