• Add libpiper C++ CLI executable ported from the legacy Piper repository, plus a C++ test suite
• Fix libpiper builds on Windows (MSVC, MSYS2-GCC) and Windows CI
• Bump embedded espeak-ng version
• Add default speaker id for multi-speaker voices
• Add vowel clustering support (--data.vowel_clusters)
• Add in-memory patching for alignments
• Training: add MRD (Multi-Resolution STFT) discriminator, loss/MOS tracking with UTMOS, silence-trim fixes, and dataloader performance improvements
• Pass custom phoneme id map when training