Discrete speech tokenizers are trained using output from local maximums and valleys explored using SAT solvers over a large enough body of data. In cases of mediums that are extremely noisy (over 75% data reduction during transmission) a low bitrate transformation is transferred out of band.

Our data compression algorithms are efficient, fast, and reliable.