Study of the Effects of Control Factors on Speech Features Using Taguchi Method

Abstract

In this paper, Taguchi optimization method is used for the first time to study the effect of some factors on the features included in speech signal. These factors are ambient temperature, speaker's age, number of words, and number of letters in each word. Taguchi method has the advantage of using a small number of experiments while getting good results for finding which factor has more effect on the phenomena under study. In this method, the number of experiments is based on the orthogonal array that depends on the number and level of the control factors. The main advantage of this method is evident in relation to reducing number of needed experiments where time, cost and material resources are minimized compared to other methods that rely mainly on trial-and-error approach. Nine representative experiments have been carried out. Each one is repeated three times. MATLAB has been employed to find the speech features under study including entropy and linear predictive coding (LPC). Minitab statistical software has been used to analyze the results based on Taguchi method. It has been found out that the order of factors according to their effect on the abovementioned speech features from higher to lower is number of words, number of letters, age, and ambient temperature. In order to validate these results, a linear regression analysis, which has confirmed the results obtained using Taguchi method, has been carried out. The results are immensely beneficial for researchers in improving speech analysis related applications by understanding which factors have more effect on speech features.