Correction method for Chinese speech synthesis tone

Abstract

The invention discloses a correction method for Chinese speech synthesis tone. According to the invention, a text analysis module receives optional text information to be synthetized; integral synthesis tagging information is outputted according to the syllable and rhythm hierarchical structure; a parameter voice synthesis module receives the synthesis tagging information of the text analysis module; synthetic voice signal is outputted through a parameter generation method of reftone; an off-line training module is responsible for the training of hidden Markov models; a reftone model is used for generating individual syllabic reference base frequency envelope; and a synthesis parameter model is used for gaining synthetic parameter sequence. The invention can solve the problem that the Chinese speech synthesis middle tone based on the hidden Markov model is unstable, thereby greatly improving the natural degree and rhythm of the synthetic speech.

Description

technical field [0001] The invention designs a parameterized speech synthesis method, and in particular relates to a tone correction method for Chinese speech synthesis. Background technique [0002] The goal of speech synthesis technology is to make electronic devices sound like humans. With the development of speech synthesis technology, the sound quality, naturalness, and intelligence of synthesized voices have been greatly improved, and the most rapid development is speech synthesis technology based on parametric statistical models. Parametric statistical speech synthesis technology based on Hidden Markov Model is a representative of this kind of method. The synthesized sound quality has high coherence and flexibility, and the required resources occupy less space, which has great practicality and research value. This method is divided into two parts, one is the offline model training part, and the other is the online speech synthesis part. In the offline training part...

AI Technical Summary

Problems solved by technology

[0003] Since the pitch accuracy of a single syllable plays a vital role in the intelligibility and naturalness of the synthesized speech in Chinese synthesis, the Hidden Markov Model belongs to a segmented model segmented by state. Each segment are independent of each other, causing the fundamental frequency within a syllable to appear uneven, resulting in a significant decline in intelligibility and naturalness

Images