Voice recognition (speech analysis using MATLAB) / Jamilah Bunanjin

Speech signal processing and analyzing is an important research. In this project the signal is processed and analyzed to determine whether it is voiced or unvoiced signal by using autocorrelation method. The data used are word 'SAYA' and 'DIA'. From word SAYA, the frames that can be produced from 11900 samples are 25 frames of data. While for word DIA, the frames that can be produced from 4500 samples are 17 frames of data with each frame (from word SA YA and DIA) uniformly having 300 samples. The length of each frame is the same. To determine whether the signal is either voiced or unvoiced is by analyzing at the peak of autocorrelation function on the error signal. If the second peak is 30% higher than the first peak, so it is declared as 'voiced' and if the peak is less than 30% from the first peak, so it is declared as 'unvoiced'. MATLAB is used to find the comparison between the input data (original data) and new data (filtered data) and also to find the peak of autocorrelation function from the signal.