طريقة جديدة لأجل تقطيع دقيق لصوت الكلمة العربية

Abstract EN

In this research we introduce a new algorithm in segmenting Arabic word sound signal to its basic phonemes that consist from it , this algorithm try to find fixed segmentation points with very high accuracy .

This algorithm depend on two type of analysis signal to find segmentation sound data in order to reach to the optimal segmenting with low error .

In the first part of this algorithm the input data is analyzed by using the Discrete Wavelet Transformation to three level , and then one node from each level is selected which is represent the first node from the low level ( approximation coefficients) , then the data of three nodes are manipulated together and the result be value negative or positive for each frame of sound ,and these resulted data is saved .

In the second part of algorithm the input data are analyzed again by Fast Fourier Transformation , then calculate two basic value of the analyzed data sound by using this method , these two value are mean of rows (mean of each frame of coefficients) and mean of columns, after these value are calculated will be used in sequence of processes that depend on the same sound coefficients and the final result be data with negative value and another with positive value .

Finally add the result of first part of algorithm (depending on wavelet transformation) with the result of second part (depending on fourier transformation) and the final result is one column with negative and positive values where the sequence of negative value represent consonant sound region and the negative represent vowel sound region ,thus complete the process of determining the segmentation points from this column in very high easy and accuracy by search in this column to find the transport point from negative value to positive value also from positive to negative because it represent segmentation points.

This work run on Pentium 4 computer by using the scientific programming language MATLAB .