Speech recognition decoding method and speech recognition decoding device

摘要: The invention discloses a speech recognition decoding method and device, which belong to the field of processing. comprises steps: information is received, acoustic features are extracted; calculated according continuous time sequence classification model; if frame in feature non null model frame, weighted finite state transducer adaptive modeling used for linguistic search historical data stored, or otherwise, discarded. Through building model, more accurate; through using transducer, representation efficient, nearly 50% computation memory resource consumption reduced; by phoneme synchronization case decoding, calculation quantity times effectively reduced.