An efficient Voice Activity Detection method for Automatic Tamil Speech Recognition System using VADSOHN algorithm

Abstract

In the emerging trend of speech technologies, speech/non-speech detection is an unsolved problem, which affects numerous speech related applications. Especially in robust speech recognition, there is always a need for a noise reduction scheme working in combination with a precise Voice Activity Detector (VAD). The approach in this paper is based on filtering Tamil speech signal, which is fed as an input to avoid high-energy noisy component and then to determine the speech/non speech in a given signal. Henceforth this research proposes an improved VAD for Automatic Tamil Speech Recognition System (ATSRS). Four types of Infinite Impulse Response (IIR) filters and three VAD techniques are used for the above-mentioned work. These techniques are compared and the best combination of filter with VAD is proposed for ATSRS. By experimental results the proposed method, which is the combination of elliptic Filter with VAD based on sohn's method, which gives better results more effectively. The performance evaluation is done with the speech quality measures like MSE and PSNR.