Improving speech identification using various digital code conditions

摘要

A microphone (M) output is fed to a communication network (KN) that has a coding unit (CD) and this connects to a speech analyser (SPA). Identified words are tagged with characteristic vectors (MV) stored in a memory (MVSP) and the values are compared with average values (MW1-MWn) in a code identifier (CE). Differences are used to switch (SM) between a number of speech identification stages (SE1-SEn). The system is trained using multiple speakers to provide word models to give improved accuracy of identification.