Time encoded signal processing and recognition with vector quantization: applied to Arabic numerals

Abstract

This article presents our contribution to speaker recognition using Arabic numerals. This recognition is based on hybridization between the time encoded signal processing and recognition (TESPAR) technique and vector quantization (VQ), in order to consolidate the classification step thanks to this combination. To set up an effective and efficient recognition system, we used a corpus recorded under ideal conditions, minimizing the differences between the reference corpus and the test corpus. We also applied the linear discriminant analysis (LDA) technique in order to discriminate the acoustic vectors and minimize the representative space. This hybridization indicated a quantifiable increase in the speaker recognition rate with the ten Arabic numerals (0–9).