🔊 JakeTTS

jaketts — also available as the shorter jtts command — is a local text-to-speech utility for macOS powered by the open-weight Kokoro-82M model via ONNX Runtime.

It can play synthesized speech directly through your speakers, save WAV files, read plain-text files, switch between Kokoro voices, adjust playback speed, and launch a Qt desktop interface when run with no arguments.

The default voice is bm_george.

JakeTTS 1.0.10 is the final release of the open-source application. The JakeTTS source code remains available under 0BSD, but the project is no longer under active feature development.

Project status

JakeTTS 1.0.10 is the final release. No new features, enhancement requests, or request-driven changes are planned. The repository is intended to become read-only after the final release is published.

If you use the free/open-source JakeTTS and want to extend it, change its behavior, add platforms, or maintain it further, fork the repository and continue from your own fork. A severe defect discovered before archival may be addressed at the maintainer's discretion, but no ongoing maintenance or support is promised.

Features

• 🔊 Direct speaker playback
• 💾 WAV file export
• 📖 Plain-text file input
• 🗣️ Voice selection from the installed Kokoro voice bundle
• ⏩ Adjustable speech speed with a 0.80× default
• 🌍 Kokoro voice families for English, Spanish, French, Hindi, Italian, Japanese, Brazilian Portuguese, and Mandarin Chinese
• 🖥️ Qt desktop GUI with voice, exact speed, volume, and Stop controls
• 🔒 Fully local synthesis once dependencies and runtime assets are installed — no model or voice downloads at runtime
• ⚡ jaketts and jtts command aliases
• 🇯🇵 Lightweight local Japanese support using UniDic Lite instead of a first-run full-UniDic download
• 🧠 Kokoro ONNX FP16 inference instead of the much larger PyTorch runtime

Runtime assets

JakeTTS 1.0.10 uses two local Kokoro assets:

• kokoro-v1.0.fp16.onnx
• voices-v1.0.bin

JakeTTS never downloads these files while synthesizing speech. It looks for them in this order:

1. the directory named by JAKETTS_ASSET_DIR
2. an assets directory beside jaketts.py
3. <python-prefix>/share/jaketts (used by the Homebrew package)

Homebrew installs the assets automatically. PyPI users can install them in a stable per-user directory and point JakeTTS at it:

ASSET_DIR="$HOME/.local/share/jaketts"
mkdir -p "$ASSET_DIR"

curl -L --fail \
  -o "$ASSET_DIR/kokoro-v1.0.fp16.onnx" \
  https://github.com/thewh1teagle/kokoro-onnx/releases/download/model-files-v1.1/kokoro-v1.0.fp16.onnx

curl -L --fail \
  -o "$ASSET_DIR/voices-v1.0.bin" \
  https://github.com/thewh1teagle/kokoro-onnx/releases/download/model-files-v1.1/voices-v1.0.bin

export JAKETTS_ASSET_DIR="$ASSET_DIR"

The expected SHA-256 hashes are:

Asset	SHA-256
kokoro-v1.0.fp16.onnx	f3a290d384fbb27966d462905c71a46cef9e5fd00516b40df32a0b4afe77ac96
voices-v1.0.bin	bca610b8308e8d99f32e6fe4197e7ec01679264efed0cac9140fe9c29f1fbf7d

For a source checkout, you may instead place both files in ./assets, which is beside jaketts.py.

Once dependencies and these two assets are installed, normal JakeTTS use does not require an internet connection.

Requirements

The Python package requires Python 3.10 through Python 3.12 (>=3.10,<3.13). The final release was tested primarily on Python 3.12, and the Homebrew package uses Python 3.12.

Version 1.0.10 uses kokoro-onnx for inference. Japanese and Chinese use lightweight local Misaki G2P paths before the phonemes are passed to the same ONNX model. Other supported languages use the local eSpeak-based phonemizer included in the installed runtime.

Installation with Homebrew

On Apple Silicon Macs running macOS 14 Sonoma or newer:

brew install ofalltrades/tap/jaketts

That installs the Python runtime, the local FP16 Kokoro model, the 54-voice v1.0 bundle, and both command aliases.

Verify the installation:

jtts -v

Launch the desktop GUI:

jtts

Or synthesize from the terminal:

jtts "Hello from JakeTTS"

The Homebrew formula is maintained at ofalltrades/homebrew-tap.

Installation from PyPI

Install the Python code and dependencies:

python -m pip install jaketts

Then install the two runtime assets using the commands in Runtime assets above. Keep JAKETTS_ASSET_DIR set to that directory when running JakeTTS.

Both commands are installed:

jaketts --version
jtts --version

Installation from GitHub

git clone https://github.com/ofalltrades/jaketts.git
cd jaketts
python3.12 -m venv .venv
source .venv/bin/activate
python -m pip install -e .

Download the two runtime assets into ./assets (using the same upstream URLs and hashes shown in Runtime assets). Editable mode means changes to jaketts.py are used immediately without reinstalling the package.

Usage

Launch the desktop GUI

jtts

The GUI launches in its own process, so the terminal prompt returns immediately while the desktop window remains open. It warms the local ONNX speech engine in a background thread.

Speak text using the default voice

jtts "Three Rings for the Elven-kings under the sky."

Choose a voice

Voice IDs can be placed before or after the text:

jtts af_sarah "Hello from an American female voice."
jtts "This uses Adam." am_adam

If no installed voice ID is supplied, bm_george is used.

Japanese and Chinese

Japanese and Mandarin Chinese are phonemized locally and then synthesized by Kokoro through ONNX Runtime:

jtts jf_alpha "こんにちは世界"
jtts zf_xiaobei "清晨的阳光从窗户照了进来。"

There is no first-run dictionary or model download.

Read a text file

jtts story.txt
jtts am_adam story.txt

Save WAV audio

A bare -o saves to output.wav:

jtts -o "Save this narration."

Specify a filename:

jtts -o narration.wav "Save this narration."
jtts --output narration.wav am_adam "Save this narration."

Equals syntax is supported:

jtts --output=narration.wav "Save this narration."

Change speech speed

The default is 0.8 (shown as 0.80× in the GUI):

jtts -s 1.25 "Speak this a little faster."
jtts --speed 0.9 am_adam "Speak this a little slower."
jtts --speed=1.1 "Slightly faster speech."

kokoro-onnx exposes sentence and clause pauses independently from spoken-word speed. JakeTTS 1.0.10 uses conservative fixed pause values internally.

Flexible argument ordering

jtts am_adam "Hello" -o hello.wav
jtts -o hello.wav "Hello" am_adam
jtts "Hello" -s 1.1 am_adam -o hello.wav
jtts am_adam "Hello" -o

Show the installed version

jtts -v
jtts --version

Voice reference

The voice menu and CLI voice recognition are read directly from the installed voices-v1.0.bin bundle rather than from a duplicated hard-coded list. The current bundle contains 54 voices.

Some useful examples include:

Voice ID	Family	Description
bm_george	British English	Default narrator
bm_lewis	British English	Male British voice
bf_emma	British English	Female British voice
af_heart	American English	Expressive female voice
af_sarah	American English	Female American voice
am_adam	American English	Male American voice
ff_siwis	French	Female French voice
jf_alpha	Japanese	Female Japanese voice
pf_dora	Brazilian Portuguese	Female Brazilian Portuguese voice
zf_xiaobei	Mandarin Chinese	Female Mandarin Chinese voice

Development and testing

The final source tree includes test_app.sh, a 26-test integration matrix. Run it from an editable install in an isolated virtual environment so the commands under test resolve to the current checkout rather than to an older global or Homebrew installation:

python3.12 -m venv .venv
source .venv/bin/activate
python -m pip install -e .

# Put the two runtime files in ./assets, or point at an existing asset directory.
export JAKETTS_ASSET_DIR="$HOME/.local/share/jaketts"

./test_app.sh

The test script intentionally refuses to run against a different installed copy of JakeTTS.

Privacy / offline behavior

JakeTTS does not use a hosted TTS API and does not fetch model or voice files during normal operation. To verify an installation, disconnect networking and synthesize or export a WAV file; all speech inference should continue to work locally.

License

JakeTTS's own source code is released under the BSD Zero Clause License (0BSD). You may use, copy, modify, distribute, or sell the JakeTTS code for any purpose, with or without fee, subject to the license text in LICENSE.

0BSD applies only to JakeTTS's own code. It does not relicense the model, Python packages, native libraries, dictionaries, or other third-party components used at runtime. The current dependency tree includes components under Apache-2.0, MIT/BSD-style licenses, LGPL, and GPL terms. In particular, kokoro-onnx depends on GPLv3-or-later phonemizer and on espeakng-loader, which loads the GPLv3-or-later eSpeak NG library, while PySide6/Qt and libsndfile have LGPL/commercial licensing considerations.

If you redistribute JakeTTS or build another product from it, you are responsible for complying with the licenses of the third-party components you distribute or link. Do not assume that JakeTTS's 0BSD license makes the complete runtime stack permissively licensed. See THIRD_PARTY_NOTICES.md in the source repository for the key licensing notes and upstream sources.