How it works

BEST uses hierarchical intermediate representations to align neural signals with audio and text, then a compact decoder maps that aligned signal directly to words.