LibCube: Find new sounds from audio synths easier (github.com)

LibCube is a new C++/C-API DLL release that uses machine learning to perform "parameter reduction" for hardware/software audio synthesizers: it maps large, hard-to-tweak parameter sets down to a 2–3 dimensional latent space so users can more easily explore and discover new sounds (think: a small, searchable “sound cube” of good presets). The package includes cube_param.dll, the cube_interface.h C header, and a test example (test_cube_c_api.cpp). The provided example uses t-SNE for dimensionality reduction (recommended) with complexity roughly O(N log N) in the number of presets, and leverages multicore CPUs for fast computation on modern machines. Technically, LibCube reduces dimensionality with t-SNE and reconstructs full parameter vectors via neural networks, which makes the search/exploration workflow practical but prevents realtime, sample-rate automation: reconstruction is CPU-intensive so you can adjust parameters and then play the instrument offline, but you cannot continuously automate parameters at 44.1 kHz. The release ships with compile scripts for MSYS2/MINGW or Visual C++ (init_env.bat required for VC++), and depends on a few Linux DLLs with their own licenses that nevertheless allow commercial use. For sound designers and ML-in-audio developers, LibCube offers an immediately usable toolchain to index, visualize and generate novel synth patches without hand-tuning dozens of knobs.