Vowel perception studies indicate that in vowel recognition, hearing uses short-term frequency analysis with the temporal window duration of the order of 10 is. This suggests that vowels should be synthesized not only in terms of their long-term amplitude spectra, but also that the proper phase relationships of their components should be preserved. For these reasons the "Wavelet Speech Synthesizer" has been developed for generation of voiced speech sounds. The signal is synthesized in the time domain as a convolution of the glottal wave with wavelets, each representing an impulse response of a formant resonantor of the vocal tract. The paper describes the principles of the synthesis, the process of encoding the speech parameter tracks by the user, and several possible classroom and research applications of this synthesizer.