Performance Analysis of Different Wavelet Families in Recognizing Speech

Automatic Speech Recognition (ASR) is one of the challenging areas of research in digital signal processing and engineering due to its wide range of applications. In this paper, a speech recognition system is developed for recognizing speaker independent spoken isolated words in Malayalam. Voice signals are sampled directly from the microphone and the features are extracted using Discrete Wavelet Transforms (DWT). Different types of wavelet families are available for speech processing and mathematical analysis. Since DWT uses wavelets, the main issue here is to find out the optimal wavelets for speech recognition. This paper investigates the performance of different wavelet families like Haar, Daubechies, Symlets, Coiflets etc. A multilayer neural network trained with back propagation algorithm is used for classification. The proposed method is implemented for 1000 speakers uttering 10 isolated words each. The experimental results show different recognition accuracies for different wavelet families and the best result of 90.2% is obtained using Daubechies wavelet families with order 4.