Speaker-Independent Connected Speech

摘要

The final report for the Speaker-Independent Connected Speech effort document's RCA's accomplishments in five general areas: (1) A two hundred word database was recorded for a group of 46 speakers across eight samplings of dialect regions; (2) A pitch synchronous acoustic parameter extraction algorithm was designed and implemented; (3) A series of human perception experiments led to several conclusions which are pertinent to automatic speech recognition; (4) A feature-based speaker-independent connected-digit recognizer was implemented; (5) The parameter extraction and recognition algorithms were tested and evaluated under a variety of noise conditions.