An ML-based UTAU resampler. Rebuilt from hifisampler with better cross-platform performance and user experience.
- uv
- Python 3.11+ (managed by uv)
git clone https://github.com/0x24a/ccbsampler.git
cd ccbsampler
uv sync
uv run setup.py
This downloads the vocoder and the hnsep model. Model files are not part of the repo, so this step is necessary.
uv run main.py
The server listens on 127.0.0.1:8572. On the first launch, the vocoder checkpoint is converted to ONNX and it might take a moment. After converting, each launch can load the model in a couple of seconds.
Note
On macOS the vocoder runs on CoreML by default. On Windows and Linux the default install runs on CPU. See Hardware acceleration to enable your GPU.
- Download the client binary for your architecture from Actions (
ccbsampler-client-<os>-<arch>.tar.gz) or Releases - Unpack it, and you will see an executable file.
- Start OpenUtau and create a project. Select any UTAU singer, then change the renderer from the default "WORLDLINE-R" to "CLASSIC". (Leave it as-is if it's already CLASSIC)
- Click on the "Track Settings" icon on the right side to open a menu. In the "Resampler" section, click "Location", then your
Resamplersfolder will show up. - Copy the executable file extracted earlier to the
Resamplersfolder. - Close and re-open the "Track Settings" menu, then you will be able to find ccbsampler-client in the dropdown.
- Optionally set it as the default resampler by clicking on "Set As Default".
- Click on "OK" to apply the settings.
- Hardware acceleration · how to get the model running on your GPU
- Configuration · explaining
config.yaml - API reference · HTTP API, render metrics, and UTAU flags
Made with ❤️ by 0x24a