The synthesis of female voices using a software synthesizer

The existing cascade-paralle1 software speech synthesizer modelled on the system of Dennis Klatt was further improved. A new, more universal periodic pulse source was added to it and the program environment of the main procedure of speech synthesis was made more useful. The usability of the improved synthesizer in the synthesis of female voices was tested. Natural vowels produced by 4 female and 2 male speakers were resynthesized and subjected to auditory evaluation by 10 listeners. The synthesizer was controlled using data derived from LPC analysis. The subjects' task was to identify the natural and synthetic vowels presented in random order, to indicate which of them sounded unnatural, to specify each speaker's gender and, in a subsequent test following familiarization with the voices, to give his or her name. The results obtained are indicative of high naturalness of the synthetic speech, reflecting personal voice features, and corroborate the usability of LPC in the extraction of parameters for the synthesis.