Repository navigation
Installation
LocalSoundsAPI is fully portable: everything lives in one folder, with its own Python and audio tools. Nothing is installed into Windows.
| OS | 64-bit Windows 10 or 11 |
| GPU | NVIDIA GPU recommended (CPU works for Kokoro and Whisper) |
| Disk | ~2.1 GB for the runtime download, plus models (≈ 2 GB for Kokoro alone, ≈ 30 GB for every engine) |
| Rights | No administrator rights, system Python or 7-Zip needed |
For reference, on a two-GPU test machine every engine was loaded at once: XTTS + Fish Speech + Kokoro + Whisper used ≈ 7 GB on an RTX 3060 (12 GB), and Stable Audio + ACE-Step + CLAP used ≈ 12 GB on an RTX 3090 (24 GB). One mid-size GPU is plenty if you load only the engines you use; see Concurrency & Progress.
-
Get the code. Clone it, or download Source code (zip) from the latest release and extract it anywhere (a short path such as
D:\LocalSoundsAPIis best).git clone https://github.com/aivrar/LocalSoundsAPI.git
-
Double-click
install.bat. Setup downloads the portable Python environment and audio tools, verifies their SHA-256 checksums, and extracts them into place. -
The Launcher opens automatically. Download the models you want from its Models & Tools tab, or let each engine download its model the first time you load it.

You can also double-click launcher.bat on a fresh clone. It notices the missing runtime and starts the installer for you.
- Finds the newest GitHub release that has both dependency assets (currently the v1.0 bundle)
- Downloads
portable-python-env-v1.7zandbin.zip, resuming if interrupted - Verifies the SHA-256 digest GitHub publishes for each file
- Extracts Python into
python/and FFmpeg, RubberBand and eSpeak-ng intobin/ - Leaves Windows and any system Python untouched
-
Open the v1.0 dependency bundle.
-
Download
portable-python-env-v1.7zand extract it into the project folder (it createspython/). -
Download
bin.zipand extract it into the same folder (it fillsbin/). -
Check these files exist:
python/python.exe bin/ffmpeg/bin/ffmpeg.exe bin/rubberband/rubberband.exe bin/espeak-ng/libespeak-ng.dll -
Double-click
launcher.bat.
Don't replace the bundled environment with
pip install -r requirements.txt. The full ML and CUDA stack is pre-assembled in the release asset.
| Way | What it does |
|---|---|
launcher.bat |
Desktop launcher: pick GPUs, run one or more instances, view logs, download models |
(portable) LocalSoundsAPI-Single.bat |
One instance on port 5006, opens your browser |
(portable) LocalSoundsAPI-Multi.bat |
Several instances on consecutive ports |
Then open http://127.0.0.1:5006 and continue with Getting Started.
git pull origin main
install.batIf the runtime is already installed, install.bat only verifies it and opens the Launcher. Use install.bat -Force to reinstall the dependency bundle on purpose.
After the runtime and model downloads, LocalSoundsAPI runs fully offline.
LocalSoundsAPI · Latest release · PDF manual · Issues · MIT licensed
Get started
Speech
Music & sound
Production
Developers
Help