Upload a record -- any audio format (wav, flac, ogg, mp3, ...), csv or numpy; up to 4 channels are used -- and synthesize new signal blocks that match its Slepian-multitaper cross-spectral density -- power spectra, coherences and cross phases -- together with its higher-order moments: skewness, kurtosis, and the joint cross-moments that govern coincident peaks. Non-stationary records can be resynthesized piecewise (multimodel): the record is chopped into consecutive block-length sections, each gets its own spectral model and one synthesized block, and the blocks are merged back into one continuous signal that tracks the record's timeline. Powered by the open-source mimoshape phase optimizer.
Merge methods: aligned crossfade rotates each next block (blocks are periodic, so this is statistically free) to best match the outgoing waveform, then applies a short equal-power fade -- best for audio. C1 splice constrains each block's head value and slope to continue the previous block during optimization: mathematically seamless, no fade zone. Zero joints pins every block start and end to zero -- the classic burst format for shaker testing.
Uploading and parsing …