Ph.D. Tezi Görüntüleme
Speech signals are the most important fact for the communication since primitive ages. Human beings transmit their feelings, thoughts and desires to the others through speech signals. These signals can be used to achieve a communication between human beings and the computers. If the computers recognize and execute the speech commands produced by the human beings, this will allow the normal and handicapped people to make use of computer technology in large extent.
There are many considerable researchs in speech recognition for the English language than the Turkish Language. Actually, the researchs for the Turkish language are not enough to develop a complete Turkish speaker independent speech recognition system. This speaker independent speech recognition system will give a quite important chance to the Turkish handicapped people in order to control electronic equipment around them.
The aim of this study is to devolop a complete speaker independent Turkish syllable recognition system. This system will be separated voiced or unvoiced letters for the input syllable, thus the input syllable will be recognized from the recognizing of each voiced or unvoiced letters.
Linear Predictive Coding, Fast Fourier Transform and Welch Averaging Periodogram methods have been used for the spectral analysis of Turkish letters. The spectral analysis results and cross correlation technique have been used to develop a speaker independent Turkish syllable recognition system. Then, the recognition system has been tested by new patterns supplied later. Finally, the obtained recognition rate has been discussed in various respects.
Keywords:Speech Analysis, Speech Recognition, Linear Predictive Coding, Cross Correlation, Fast Fourier Transform, Periodogram, Speech Detection.