Class-Discriminative Weighted Distortion Measure for VQ-based Speaker Identification

Abstract

Abstract. We consider the distortion measure in vector quantization based speaker identification system. The model of a speaker is a codebook generated from the set of feature vectors from the speakers voice sample. The matching is performed by evaluating the distortions between the unknown speech sample and the models in the speaker database. In this paper, we introduce a weighted distortion measure that takes into account the correlations between the known models in the database. Larger weights are assigned to vectors that have high discriminating power between the speakers and vice versa.

Principal component based classification for text-independent speaker identification

“…Further increase on the value of r (16,17,18,19), produce no significant improvement. It is seen that when the principal component space dimension, r, increases, the truncation error space dimension decreases as expected, resulting in a decrease in the identification rate.…”

“…LBG algorithm is a widely used VQ algorithm in speaker identification [15], [16]. Table I shows the identification rates using VQ classifier with different codebook sizes for TIMIT and NTIMIT databases.…”

The Improved VQ Algorithm for Speaker Recognition

“…The improved VQ algorithms in the speaker recognition system has mainly the following two aspects: the use of space vector correlation improve algorithm: LVQ, GVQ [2] and the weighting method to distinguish between categories (CDWM) [3] , and the use of space vector the relevance of improved algorithm , such as PNDM [4] . In this paper, in order to increase the distinction between the worlds of words and effectively improve of recognition rate, which we improve and achieve by distinguishing between types of PNDM and combined weight of the algorithm.…”

A Speaker Pruning Algorithm for Real-Time Speaker Identification

“…All speakers are modeled by a codebook [4,7] of 64 vectors using the generalized Lloyd algorithm (GLA) as the clustering method [5]. The pattern comparison method is the average quantization error (or distortion) D(X, C) between the test vector sequence X and the codebook C [7].…”