Making Reference Vectors for Speaker Identification Using the Combination of Dp-Matching and Lvq Clustering Technique
Munehiro Namba, Yoshiteru Ishida · Information Sciences, Signal Processing and their Applications · 1996
This paper presents a method for making reference vectors in speaker recognition. The DP-matching technique is incorporated into the LVQ neural network so that the generalization of the reference vectors is possible in both time and frequency domain. Because personal characteristics are widely distributed in the feature vector space, it is usually hard to make the appropriate reference vectors in the speaker recognizer. With our technique, the recognizer becomes insensible to both the rime and spectral variations.