Title:
SPEECH SYNTHESIS BASED ON WAVEFORMS

Authors:
Chih-Hsu Hsu, Taiwan

Abstract:

In this paper, we discuss a type of synthesis by compilation of speech waveforms. In order to synthesize any articles, we need control speech waveforms with pitch patterns. We proposed a novel method, which were features of the fourth tone with wide pitch frequency, to obtain the various pitch periods of natural speech waveforms, instead of interpolation or decimation of natural speech waveforms. Thus, we took a pitch period of natural speech waveforms as a basic unit, combining with synthesis rules of rhythm of natural speech, to synthesize sentences. In order to synthesize a complete sentence, we divided it into four steps. The first step of synthesis is a word, the second is a phrase, the third is an inspiration, finally synthesizes a complete sentence. Synthesis of words includes (i) vowels, (ii) diphthongs, (iii) consonant combined with vowels. The processing of vowels, we utilized the features of the fourth tone with wide range of pitch frequency, and retained all of the periodical speech waveforms for basically syntactic units. The processing of diphthongs, we preserved the parts of acutely changed waveforms in the middle, then processed others as vowels. The processing of consonant combined with vowels, we need retain consonants and a little parts of adjacent waveform of vowels, then imitate others as the above two classes. The results of listening tests showed that it could really synthesize sentences with fluency.