This repository contains prebuilt wheels for Misaki TTS and its dependencies, separated by language and CPU/GPU support to optimize container size and build time.
Misaki supports the following languages, each available as a separate build:
en- Englishja- Japaneseko- Koreanzh- Chinesevi- Vietnameseall- All languages combined
Each language pack can be built with or without GPU support:
cpu- CPU-only build (smaller, no NVIDIA dependencies)gpu- With CUDA/GPU support (includes NVIDIA libraries)
Each build is available as both:
- A GitHub release with wheel files
- A Docker image with prebuilt wheels
# For CPU-only English build (smallest possible image)
COPY --from=ghcr.io/remsky/prebuilt_tts_wheels/misaki-deps:0.9.4-en-cpu-py3.10 /wheelhouse /wheelhouse
RUN pip install --no-index --find-links=/wheelhouse misaki[en]==0.9.4
# For GPU-enabled all-languages build
COPY --from=ghcr.io/remsky/prebuilt_tts_wheels/misaki-deps:0.9.4-all-gpu-py3.10 /wheelhouse /wheelhouse
RUN pip install --no-index --find-links=/wheelhouse misaki[en,ja,ko,zh,vi]==0.9.4To build wheels for a specific language and hardware combination:
- Go to the "Actions" tab in this repository
- Select the "Build Misaki Dependencies Per Language" workflow
- Click "Run workflow"
- Enter the Misaki version you want to build
- Select the language from the dropdown menu
- Select either "cpu" or "gpu" for hardware support
- Click "Run workflow"
- Reduced Docker Image Size:
- CPU-only builds exclude large NVIDIA CUDA libraries (over 1.5GB)
- Single language builds exclude unnecessary language models
- Faster Builds: Building for a specific language/hardware combination is much faster
- Targeted Updates: You can update just one language pack or hardware variant when needed
- Efficient Deployment: Use exactly what you need in production
All images are available on GitHub Container Registry at:
ghcr.io/remsky/prebuilt_tts_wheels/misaki-deps