A PITCH BASED VAD ADOPTING QUASI-ANSI 1/3 OCTAVE FILTER BANK WITH 11.3 ms LATENCY FOR MONOSYLLABLE HEARING AIDS

Abstract

This paper presents a pitch based voice activity detection (PBVAD) algorithm adopting a quasi-ANSI 1/3 octave filter bank which has low group delay for realistic implementation in hearing aids systems. For compensating the drawback of low resolution resulted from quasi-ASNI filter bank, this pitch based VAD algorithm integrals the features of monosyllable speech such as pitch and corresponding harmonics, onset and time of word length. Simulation results reveal that with more harmonics detection, the accuracy of the proposed PBVAD algorithm improves from 78.9% to 87.7%. Additionally, the proposed VAD algorithm is implemented in ANSI filter bank for comparisons. With the integration of features, the result shows the proposed algorithm can achieve similar VAD accuracy, less than 2.5%, in quasi-ANSI filter bank and ANSI filter bank. Thus, the proposed algorithm can tackle the drawback of quasi-ANSI filter bank and is also suitable for ANSI filter bank. Moreover, the latency incurred by quasi-ANSI filter bank and the proposed VAD algorithm is 11.3ms and this satisfies the requirement of HA systems for practical implementation.