Speaker recognition system.

摘要

A speaker recognition system (100) for recognizing a speaker on the basis of input speech by the use of a neural network (120) provided with a plurality of layers (120...122). The variation with time of an inherent characteristic of speech is used as an input to the neural network. Specifically, the variation with time of an average inherent characteristic of speech within a fixed time is used as the input to the neutral network. The inherent characteristics of speech used include a frequency characteristic of speech, a pitch frequency, a frequency characteristic of the speech waveform subjected to high-frequency emphasis; a linear prediction coefficient, and a PARCOR coefficient.