A new system for authentication of voice

Presents a survey of an automatic speaker verification system which is suited for different environmental conditions. The structure of the system is explained and results are discussed. The verification is accomplished by dialing a code number and speaking a pre-arranged code sentence into a customary telephone. Instructions and responses to the speakers are yielded with a voice-response system. Feature sets used for classification are selected short-time spectra, long-time spectrum, intensity contour, and stationarity contour. Results of recognition experiments show that by combination of these features a high performance is assured.