Tez özetleri Astronomi ve Uzay Bilimleri Anabilim Dalı

Performance Analysis of Speech Compression Techniques

Speech compression is defined as a process of coding speech with minimum number of bits by speech coding methods. In the literature, speech compression and speech coding concepts are used synonymously. In this thesis, one of the lossy speech compression methods LPC and recent method Wavelet Technique are examined. Also in the thesis, for the speech quality assessment modeling of the national fixed, mobile and UMTH networks to be used by the only national regulatory foundation Information Technologies and Communications Authority (BTK), a real application is done and assessments are also included.

Linear Predictive Coding (LPC) is especially used in mobile applications. In predictive coding, both the transmitter and receiver stores the previous samples of the signal. The transmitter, instead of sending the samples, only sends the prediction error that is the difference of signal and the predicted values. The receiver recovers the signal by adding prediction error to the predicted values. LPC based coders are examined in detail.

Wavelet Technique is another lossy speech compression method, which has wide range of applications. One of them is data compression. Wavelet analysis has the capability to show the trends and high level discontinuities neglected by other signal analysis methods. Wavelet analysis can compress and denoise the signal with negligible attenuation. Wavelet compression will be examined as an alternative to LPC coders in the thesis. Reference speeches are compressed and assessed.

In the literature there are many measures for speech quality. These measures are mainly divided into mathematical based objective methods and perceptual methods that models human hearing. In the thesis, assessments are done by objective measures, perceptual measures and also graphical measures.