Fundamental synthesis parameter generation method and system in singing synthesis system

Abstract

The invention discloses a fundamental synthesis parameter generation method and system in a singing synthesis system. The method comprises the following steps: acquiring a numbered musical notation to be synthesized; determining all basic synthesis units of the numbered musical notation; acquiring fundamental synthesis parameters of all the basic synthesis units according to annotations of the numbered musical notation and taking the acquired fundamental synthesis parameters as first fusion parameters; determining fundamental synthesis models corresponding to the basic synthesis units according to annotations of the numbered musical notation to obtain a fundamental synthesis model sequence; and performing fusion optimization by use of the first fusion parameters and the fundamental synthesis model sequence to acquire fundamental synthesis parameters of the basic synthesis units. By adoption of the method and system, a synthesized speech is prevented from obvious out-of-tune on the premise of maintaining relatively high similarity and expression.

Description

technical field [0001] The invention relates to the technical field of speech synthesis, in particular to a method and system for generating fundamental frequency synthesis parameters in a singing synthesis system. Background technique [0002] In recent years, singing synthesis systems have been extensively researched and applied, among which parametric synthesis systems based on Hidden Markov Model (HMM) and waveform splicing synthesis systems have reached practical standards. Compared with the synthesis system based on waveform splicing, the parametric synthesis method based on HMM requires less training data and a smaller system scale, so it has better practical value. [0003] The singing synthesis system based on parametric synthesis first determines the duration parameter sequence, fundamental frequency parameter sequence, and spectrum parameter sequence of each basic synthesis unit (such as syllable, phoneme, etc.), and then uses the parametric synthesizer to obtain ...

AI Technical Summary

Problems solved by technology

[0009] The fundamental frequency parameter generation method based on the numbered notation notation, because the marked pitch of the numbered notation does not necessarily meet the actual fundamental frequency distribution characteristics of the speaker, which leads to the inconsistency between the fundamental frequency parameters obtained according to the standard pitch and the speaker's own pronunciation characteristics, and finally The synthesized voice cannot reflect the characteristics of the speaker, the similarity is very low, and the synthesized voice is slightly dull

The parameter generation based on the fundamental frequency synthesis model obtains the fundamental frequency model that conforms to the characteristics of the speaker's timbre through the statistical simulation of real data. However, due to the loss caused by the modeling itself and the incomplete accuracy of the fundamental frequency model selection, this method cannot It is difficult to guarantee the stability of the final composite effect, and there may be obvious out-of-tune phenomena

Images