Natural text-to-speech that runs entirely on your computer.
No cloud. No subscription. No API key.
Download for Mac · Download for Windows · Website · Command line
Free and open source · macOS 14+ Apple Silicon · Windows x64 preview
Atten is a local voice studio powered by the bundled Kokoro 82M model. Your text and generated audio stay on your computer, and the complete engine works offline after installation.
- Create MP3 or WAV audio with 37 voices
- Preview voices, speed, format, and acceleration settings in the Playground
- Keep projects and exports organized in one native app
- Import text, regenerate previous work, and export anywhere
- Use the compatible CLI for scripts and automation
- Download the latest DMG.
- Open it and drag Atten into Applications.
- Launch Atten and start generating speech—everything required is included.
Atten currently requires an Apple Silicon Mac running macOS 14 or newer. The release is ad-hoc signed and not yet notarized. If macOS blocks the first launch, Control-click Atten, choose Open, then confirm. See the installation notes for the alternative Privacy & Security flow and download verification.
- Download the Windows x64 preview.
- Extract the ZIP.
- Run Atten.Windows.exe.
The Windows package uses CPU execution and works on any supported x64 Windows
10/11 machine. CUDA-aware backend support is available in the codebase and can
be packaged with scripts/build-windows.ps1 -BackendFlavor cuda, but public
release automation currently publishes the CPU preview package. The Windows app
is a preview while installer signing, MSIX packaging, and UI parity work
continue.
The original local commands remain available for automation:
bin/tts "Living the dream"
bin/tts -f README.md -v bf_emma -s 1.1 --format wav --play
bin/tts "Hello" --filename greeting --silent
bin/play --latestMachine-readable output is supported too:
bin/tts --list-voices --json
bin/tts --backend-info --device auto --json
bin/tts "Hello" --device cpu --json
bin/tts "Hello" --device auto --json--device auto selects Metal/MPS on macOS when available, CUDA on Windows/Linux
when PyTorch can use it, and CPU otherwise. Explicit --device cuda and
--device mps fail loudly if the requested accelerator is unavailable.
Run bin/tts --help for every option, or browse the voice catalog.
You will need an Apple Silicon Mac with macOS 14+, Xcode 16 with Swift 6,
Python 3.12, and uv.
bin/setup-macos
bin/attenThe development setup downloads roughly 350 MB of model weights. Published DMGs already include the pinned model and do not require Python or Homebrew.
Build and open a development app bundle:
scripts/build-app-macos
open .build/Atten.appOr work on the Swift package directly:
swift build
swift run AttenThe development app uses the repository's backend environment. To use a
backend elsewhere, set ATTEN_BACKEND_ROOT=/path/to/offline-tts.
You will need Windows 10 1809+ or Windows 11 x64, Visual Studio 2022 with
Windows App SDK tooling, .NET 10 SDK, Python 3.12, uv, and PyInstaller.
uv sync --frozen --group release --no-editable
dotnet build apps/windows/Atten.Windows/Atten.Windows.csproj -c Debug -r win-x64Build a Windows preview package on Windows:
scripts/build-windows.ps1 -BackendFlavor cpu
scripts/build-windows.ps1 -BackendFlavor cudaSee Windows port notes for packaging status and remaining production-release work.
swift test
python3 -m unittest discover -s tests -v
swift build -c release- Synthesis is local and requires no account, credentials, or internet access.
- macOS project metadata lives in
~/Library/Application Support/Atten. - Windows project metadata lives in
%LOCALAPPDATA%\Atten. - Generated audio defaults to the
Exportsfolder inside the platform app-data directory. - Existing audio under the repository's
outputs/folder is discovered in place.
See the architecture notes for implementation details and deployment guide for release packaging and verification.
Atten is available under the GNU GPL v3 or later.