In this paper, we introduce the Descend-Delta-Mean technique for feature extraction of speech signal isolated THAI digit speech. Descend-Delta-Mean (DDM) is a new method of finding the average value of a descendant. The purpose is to find the characteristic coefficients that represent the speech and more resistance to noise than another methods. We determine the mean of the difference of magnitude in order to estimate the strength of the power spectrum. The procedure consists of 3 steps as: First, the method of rearranging the strength of the spectral power is reduced to a minimum in each frequency band of the speech signal. Second, to find the difference in the arrangement. Third, to determine the value of mean from second step. The simulation results of the DDM method and the result of the experiment can recognize the isolated THAI digits speech from male and female THAI speaker with additive random white Gaussian noise to replace the dataset of testing. To find out, how this approach can be noise robustness and how much accuracy is required to distinguish data.
Paper
Full text
Descend-Delta-Mean Algorithm for Feature Extraction of Isolated THAI Digit Speech
Semantic Scholar · Computer Science · 2019
Abstract
In this paper, we introduce the Descend-Delta-Mean technique for feature extraction of speech signal isolated THAI digit speech. Descend-Delta-Mean (DDM) is a new method of finding the average value of a descendant. The purpose is to find the characteristic coefficients that represent the speech and more resistance to noise than another methods. We determine the mean of the difference of magnitude in order to estimate the strength of the power spectrum. The procedure consists of 3 steps as: First, the method of rearranging the strength of the spectral power is reduced to a minimum in each frequency band of the speech signal. Second, to find the difference in the arrangement. Third, to determine the value of mean from second step. The simulation results of the DDM method and the result of the experiment can recognize the isolated THAI digits speech from male and female THAI speaker with additive random white Gaussian noise to replace the dataset of testing. To find out, how this approach can be noise robustness and how much accuracy is required to distinguish data.