Thesis (M.Sc.) - Cairo University - Faculty Of Computers and Information - Department Of Information Technology

In this thesis we proposed modifications to the baseline system to improve its quality First we implemented a mixed - excitation model to better model the mixedexcitation nature of speech , where speech is divided into a number of bands , and voicing of each band is estimated and modelled in the HMM models Moreover a different spectral envelop representation is used , where a bigger number of spectral envelop samples are used directly to train the HMM models The spectral envelop is estimated taking into consideation the harmonic nature of the voiced speech In synthesis harmonic and noise like speech are generated and mixed according to the voicing parametersExperimental results show that the system implementing the proposed modifications produced a higher quality speech , giving higher scores in subjective listenening tests

Issued also as CD