|
In this thesis, a text-to-speech system is designed and
implemented on MS-Windows operating system. The 408 first-tone
Mandarin syllables are adopted as the synthesis units. For the
synthesis of syllable-signal, a time-domain processing method
called "Time Proportionated Interpolation of Pitch Waveform
(TPIPW)" is proposed. About the prosodic processing unit, a
rule-based method proposed by other researchers is adopted and
slightly modified here. In our method, the two parts of a
syllable, i.e. the unvoiced part (e.g. voiceless consonants)
and voiced parts (e.g. voiced consonants and vowels), are
processed separately. The name of our method is just selected
to reflect the voiced-part''s processing. By using this method,
a syllable''s tone(or pitch-contour), duration, and formant-
frequency height can be almost independently controlled.
Especially, the duration of a syllable can be more freely
changed to a value between one half and double of the original
length without notable side-effects on the other two control
factors. Besides, the function of increasing or decreasing
formant-frequency values is provided to simulate the adjusting
of vocal-track length such that the original recorded male
voice can be more naturally converted to a female''s voice. For
the unvoiced part, signal waveforms are classified into two
classes and a method is proposed to process each class
differently. This method not only synthesizes clear and
intelligible signals but also support the control of duration.
|