- Added support Phonetisaurus and Sequitur G2P to generate OOV words automatically
- Added G2P models for 'en' and 'de' trained on Zamia base models
- Added train scripts for new G2P models
- Added support for JSGF grammar language models
- Added simple Python script to optimize LM corpus before adaptation
- Many improvements to install, build and adapt code
- Reworked API of adapt script to support all new features
- Reworked completely to support SEPIA STT Server v2 (https://github.com/SEPIA-Framework/sepia-stt-server)
- Simplified code and removed several dependencies or replaced with smaller alternatives
- Added pre-built Kaldi and KenLM files and binaries (https://github.com/synesthesiam/prebuilt-apps/releases) ty to synesthesiam
- Works on x86 64Bit, ARM 32Bit and ARM 64Bit platforms
- Officially supports Raspberry Pi 4