Skrivi TTS 0.1.0

First maintained Windows release of Skrivi TTS, continuing the existing Norwegian shootout and VoxCPM2 Reader work in a separate repository.

Download the Windows ZIP, extract it, and double-click INSTALL.bat. The application appears in the Start menu and on the desktop. No separate Python installation is needed.

• Persistent model library: application updates preserve downloaded weights, presets, reference voices and saved audio.
• Four working CPU models: public VoxCPM2 Q4, Chatterbox Multilingual Q4 ONNX, Piper NVCC and Piper Talesyntese.
• Model-specific inference controls, reference WAV support, named presets and listening ratings.
• Models stay loaded between readings; results distinguish cold/warm runs and record exact settings and timings.
• Complete native WAV playback and cancellation of the owned engine process tree.

The ZIP is approximately 104 MB and contains the application and reusable runtimes, not model weights. Import existing model files or download a chosen model once from the Model library tab. All four models were imported from existing caches on the development PC without redownloading them.

Validation: five library/configuration tests; real packaged-runtime cold/warm synthesis for all four models; native WAV validation and cancellation; real reinstall preserved all 19 model/registry files. Verified on Ryzen 7 7700 / 32 GB RAM. GPU inference is not implemented. The existing Vox adapter retains its 4,096-character limit. No claim of a production-ready accessibility reader or model-weight training is made.

See README.md and PROJECT_STATUS.md for settings, provenance, build instructions and next experiments. Checksums are attached.