kittentts-rs

A Rust port of KittenTTS — an ultra-lightweight, CPU-only text-to-speech engine based on ONNX models.

Screenshots

iOS	Android

Features

• ONNX Runtime inference — uses ort (ORT 2.0 bindings) for fast CPU inference
• Full text preprocessing — numbers, currencies, abbreviations, ordinals, units, etc. → spoken words
• espeak-ng phonemisation — IPA output via the libespeak-ng C library (FFI, not subprocess)
• Same ONNX models — works with all KittenTTS HuggingFace checkpoints
• Automatic chunking — long texts split into ≤ 400-char sentence chunks, then concatenated
• Cross-platform — macOS, Linux, Windows (MSVC + MinGW), iOS, Android
• Cross-compilation — cargo cross supported out-of-the-box for Linux GNU/musl, Android, and more

Prerequisites

The espeak Cargo feature requires libespeak-ng to be installed (the shared or static library, not just the command-line tool):

# macOS
brew install espeak-ng

# Debian / Ubuntu (installs libespeak-ng.so + headers)
sudo apt install libespeak-ng-dev

# Alpine Linux (installs libespeak-ng.a for static linking)
apk add espeak-ng-dev espeak-ng-static

# Fedora / RHEL
sudo dnf install espeak-ng-devel

# Arch Linux
sudo pacman -S espeak-ng

# Windows (MSVC) — choose one:
#   Option A: official installer → https://github.com/espeak-ng/espeak-ng/releases
#   Option B: vcpkg
vcpkg install espeak-ng:x64-windows-static
#   Option C: MSYS2/MinGW64
pacman -S mingw-w64-x86_64-espeak-ng

The espeak feature is opt-in. Without it every API that accepts raw IPA input still works; only text-to-IPA conversion (and therefore the high-level generate / generate_to_file functions) is unavailable.

Installation

# Cargo.toml

# Without espeak-ng (IPA input only — no native library needed)
[dependencies]
kittentts = "0.2"

# With espeak-ng (full text input)
[dependencies]
kittentts = { version = "0.2", features = ["espeak"] }

Or add it with cargo:

cargo add kittentts
cargo add kittentts --features espeak

From GitHub

[dependencies]
kittentts = { git = "https://github.com/eugenehp/kittentts-rs", tag = "v0.2.4" }
kittentts = { git = "https://github.com/eugenehp/kittentts-rs", branch = "main" }

Quick Start

use kittentts::download;
use std::path::Path;

fn main() -> anyhow::Result<()> {
    // Downloads model from HuggingFace (cached after first run)
    let tts = download::load_from_hub("KittenML/kitten-tts-mini-0.8")?;

    println!("Available voices: {:?}", tts.available_voices);

    // Generate and save as WAV (24 kHz, float32, mono)
    tts.generate_to_file(
        "Hello from Rust! This high quality TTS model works without a GPU.",
        Path::new("output.wav"),
        "Jasper",
        1.0,   // speed (1.0 = normal)
        true,  // clean_text: run number/abbreviation expansion
    )?;

    Ok(())
}

Run the bundled example:

cargo run --example basic --features espeak
cargo run --example basic --features espeak -- --voice Luna --text "Hello world" --output hello.wav

Available Models

Available Voices (v0.8)

Bella, Jasper, Luna, Bruno, Rosie, Hugo, Kiki, Leo

API

// Load from HuggingFace Hub
let tts = kittentts::download::load_from_hub("KittenML/kitten-tts-mini-0.8")?;

// Load from local files
let tts = kittentts::model::KittenTtsOnnx::load(
    Path::new("model.onnx"),
    Path::new("voices.npz"),
    Default::default(), // speed_priors
    Default::default(), // voice_aliases
)?;

// Generate audio → Vec<f32> at 24 kHz
let audio: Vec<f32> = tts.generate("Hello!", "Jasper", 1.0, true)?;

// Generate and save to WAV
tts.generate_to_file("Hello!", Path::new("out.wav"), "Jasper", 1.0, true)?;

// Available voices
println!("{:?}", tts.available_voices);

Build Configuration

Environment Variables

Auto-build Script

Point ESPEAK_BUILD_SCRIPT at one of the provided scripts to have the build system compile libespeak-ng automatically when the archive is missing:

# macOS / Linux — any target
ESPEAK_LIB_DIR=$PWD/espeak-static/lib \
ESPEAK_BUILD_SCRIPT=$PWD/scripts/build-espeak-static.sh \
  cargo build --features espeak

# Windows
$env:ESPEAK_LIB_DIR = "$PWD\espeak-static\lib"
$env:ESPEAK_BUILD_SCRIPT = "$PWD\scripts\build-espeak-static.ps1"
cargo build --features espeak

Cross-Compilation

Linux → Linux aarch64 (Debian/Ubuntu multiarch — simplest)

sudo dpkg --add-architecture arm64
sudo apt update
sudo apt install gcc-aarch64-linux-gnu libespeak-ng-dev:arm64
rustup target add aarch64-unknown-linux-gnu
cargo build --target aarch64-unknown-linux-gnu --features espeak

Any host → Linux x86_64 or Windows x64 with cargo-zigbuild

cargo-zigbuild uses the Zig compiler as a drop-in cross-linker — no Docker, no SDK, no sysroot required for the default (no-espeak) feature set.

cargo install cargo-zigbuild
rustup target add x86_64-unknown-linux-gnu x86_64-pc-windows-gnu

# Linux x86_64 — ORT static lib downloaded automatically
cargo zigbuild --target x86_64-unknown-linux-gnu

# Windows x64 — requires llvm-dlltool to generate an ORT import library
#   (one-time setup; replace /path/to/ort with your ORT Windows DLL directory)
ORT_LIB_LOCATION=/path/to/ort \
ORT_PREFER_DYNAMIC_LINK=1 \
  cargo zigbuild --target x86_64-pc-windows-gnu

The Windows build produces a libkittentts.a containing genuine x86-64 COFF objects that link against onnxruntime.dll at runtime.

Any host → any target with cargo cross

cargo install cross --git https://github.com/cross-rs/cross

# cross reads Cross.toml and installs libespeak-ng-dev inside the container
cross build --target aarch64-unknown-linux-gnu  --features espeak
cross build --target armv7-unknown-linux-gnueabihf --features espeak
cross build --target x86_64-unknown-linux-musl  --features espeak
cross build --target riscv64gc-unknown-linux-gnu --features espeak

Cross.toml in the repository root configures pre-build commands for every supported GNU and musl target. No manual setup beyond installing cross is required.

Custom sysroot

ESPEAK_SYSROOT=/path/to/target-sysroot \
  cargo build --target aarch64-unknown-linux-gnu --features espeak

Build espeak-ng from source for a specific target

# Linux → aarch64 (requires: apt install gcc-aarch64-linux-gnu)
ESPEAK_LIB_DIR=$PWD/espeak-static/aarch64/lib \
ESPEAK_TARGET=aarch64-unknown-linux-gnu \
  bash scripts/build-espeak-static.sh

# Linux → Android arm64 (requires: ANDROID_NDK_HOME set)
ESPEAK_LIB_DIR=$PWD/espeak-static/android/lib \
ESPEAK_TARGET=aarch64-linux-android \
ANDROID_NDK_HOME=/path/to/ndk \
  bash scripts/build-espeak-static.sh

# Windows (MSYS2 or MSVC)
$env:ESPEAK_LIB_DIR = "$PWD\espeak-static\lib"
powershell -ExecutionPolicy Bypass -File scripts\build-espeak-static.ps1

iOS

bash ios/build_rust_ios.sh

Builds espeak-ng for both device (arm64) and Simulator (arm64-sim), compiles kittentts-rs for each slice, and packages everything into ios/KittenTTS.xcframework.

Android

export ANDROID_NDK_HOME=/path/to/ndk
bash android/build_rust_android.sh

Builds espeak-ng as a shared library, compiles the Rust static lib and JNI bridge, and copies all .so files into android/KittenTTSApp/app/src/main/jniLibs/arm64-v8a/.

Architecture

Input text
    ↓  TextPreprocessor  (preprocess.rs)
       • numbers / currency / percentages / ordinals → words
       • contractions, units, scientific notation, fractions, …
    ↓  chunk_text()  (model.rs)
       • split into ≤ 400-char sentence chunks
    ↓  libespeak-ng FFI  (phonemize.rs)
       • text → IPA phoneme string (en-us, with stress)
       • requires `espeak` feature + libespeak-ng linked at build time
    ↓  ipa_to_ids()  (tokenize.rs)
       • IPA chars → integer token IDs  (fixed vocab, same as Python)
       • prepend/append pad token 0
    ↓  ONNX Runtime inference  (model.rs)
       • inputs:  input_ids [1, T], style [1, D], speed [1]
       • output:  audio waveform [samples]
    ↓  tail-trim (–5 000 samples) + chunk concatenation
    ↓  Vec<f32> @ 24 kHz  or  WAV file

Crate Structure

Running Tests

# All unit tests (no native library required — 20 tests)
cargo test

# All unit + espeak-ng phonemisation tests (requires libespeak-ng — 28 tests)
cargo test --features espeak

# Integration and e2e tests using bundled model files (32 tests)
cargo test --test integration_tests

# Integration + espeak + full inference e2e tests (40 tests)
cargo test --test integration_tests --features espeak

# Point at a custom model directory
KITTENTTS_MODEL_DIR=/path/to/models cargo test --test integration_tests

# Check that all code compiles for the host target
cargo check
cargo check --features espeak

Test counts at a glance

Citation

@software{kittentts_rs_2026,
  author    = {Eugene Hauptmann},
  title     = {kittentts-rs: A Rust Port of KittenTTS},
  year      = {2026},
  url       = {https://github.com/eugenehp/kittentts-rs},
  note      = {Ultra-lightweight, CPU-only text-to-speech engine based on ONNX models}
}

@software{kittentts_2024,
  author    = {KittenML},
  title     = {KittenTTS},
  year      = {2024},
  url       = {https://github.com/eugenehp/KittenTTS}
}

Changelog

See CHANGELOG.md for a full history of releases and changes.

License

This project is licensed under the Apache License 2.0.