Authors: Haamid M. Gazi, Omar Farooq, Yusuf U. Khan, Sekharjit Datta
Addresses: Tata Consultancy Services, Air-India Building, 11th Floor Nariman Point, Mumbai, India. ' Department of Electronics Engineering, AMU Aligarh, India. ' Department of Electrical Engineering, AMU Aligarh, India. ' Applied Signal Processing Group, Department of Electronic and Electrical Engineering, Loughborough University, UK
Abstract: In this paper, Admissible Wavelet Packet (AWP)-based features are proposed for the recognition of isolated Hindi digit. AWPs are used to design a set of filter banks to follow the Mel scale. Finally, a Hidden Markov Model (HMM) is developed for recognition. The database used was collected from 48 subjects with a repetition of each digit five times. Features based on both Linear Predictive Coefficients (LPCs) as well as Mel Frequency Cepstral Coefficients (MFCCs) were extracted and their performance compared with the AWP-based features. It was found that the recognition performance using AWP-based features was superior when compared with LPC- and MFCC-based features.
Keywords: feature extraction; hidden Markov model; HMM; speech recognition; wavelet transform; speaker-independent Hindi digit recognition; isolated Hindi digit; admissible wavelet packet; AWP.
International Journal of Information and Communication Technology, 2008 Vol.1 No.2, pp.185 - 198
Published online: 28 Jun 2008 *Full-text access for editors Access for subscribers Purchase this article Comment on this article