Automatic Recognition of Speakers by Computers

摘要

The report describes the development of an improved system for the automatic recognition of speakers voices by telephone. Details are given of a system which permits a flexible combination of parameter extraction and classification procedures, and the best combinations are enumerated. A second system is described which employs a code sentence and procedures for eliminating frequency degradation by the telephone. The report includes discussions of the limitations and potentials of automatic speaker recognition, spectral preprocessing of speech signals, speech parameter contours and their time normalization, and feature extraction. The work opens the way for systems controlling authorized access to money and information by telephone.