Construction method of cross-syllable Chinese speech synthesis element with spectrum stable boundary

Abstract

The invention discloses a construction method of a cross-syllable Chinese speech synthesis element with a spectrum stable boundary, and belongs to the field of speech processing. Element segmentation on speech flow data starts from the center vowel of a syllable and ends at the center vowel of an adjacent syllable, and a cross-syllable element obtained from the segmentation is composed of two parts, i.e., the part of the center vowel of the final of a previous syllable and thereafter, and the part of the center vowel of the final of a current syllable and therebefore so as to obtain the cross-syllable element. According to the invention, by using such a method, coarticulation in a syllable and between syllables can be maintained, enormous splicing caused by a too short element can be prevented, the naturalness and continuity of synthetic speech are effectively improved, and the timbre performance of the synthetic speech is not affected.

Description

technical field [0001] The invention relates to a definition and construction method of a Chinese speech synthesis primitive, in particular to an automatic construction method of a cross-syllable Chinese speech synthesis primitive with spectrum stability boundaries. It belongs to the field of speech processing. Background technique [0002] The selection of speech synthesis primitives is a crucial link in speech synthesis. Reasonable selection of speech primitives and construction of primitive databases are of great significance to speech synthesis. There is no uniform and absolute evaluation standard for the selection of primitives, and it will be limited by conditions such as language, application field, training data volume, and storage requirements. Common speech synthesis primitives include phonemes, diphones, triphones, demisyllables, syllables, words, etc. (see Taylor, Paul. Text-to-speech synthesis. Cambridge University Press, 2009). The selection of these synthet...

AI Technical Summary

Problems solved by technology

For the former, the synthesis primitive is usually a diphone, which generally contains two phonemes, and its boundary is a stable segment in the two phonemes. Although this method also takes into account the influence of co-pronunciation, the structure of diphones leads to There are a large number of splicing points in synthetic speech, which not only increases the complexity of the algorithm, but also easily leads to unnatural cohesion; for the latter, syllables are usually selected as the synthesis primitive, which ensures the internal coherence of syllables, but the synergy between syllables Pronunciation is broken

Images