Speech synthesis method for generating new tone

Abstract

The invention discloses a speech synthesis method for generating a new tone. The method comprises the following steps of training a deep neural network by using a plurality of sound base data to forma first synthesis model; respectively training the first synthesis model by using the plurality of sound base data to form a plurality of second synthesis models corresponding to the plurality of sound base data; reasoning a first output parameter by using the first synthesis model; reasoning a plurality of second output parameters corresponding to the second synthesis models by using the plurality of second synthesis models to form a second output parameter group; performing weighted stacking on the second output parameter group to form acoustic parameters; and reconstructing the acoustic parameters by a vocoder to form synthesis speech. The method provided by the invention has the advantages that the synthesis of the speech with the new tone can be realized under the condition of not making a new sound base; and the tone of the synthesized speech can be flexibly modulated according to a synthesis model corresponding to the existing sound producer sound base data, the synthesis efficiency is not obviously changed, and the problems of complicated work procedure, long period and high cost of the manufacturing of the new sound producer sound base are solved.

Description

technical field

[0001] The invention relates to the field of speech synthesis, in particular to a speech synthesis method for generating new timbres. Background technique

[0002] Speech synthesis, also known as text-to-speech (Text to Speech) technology, refers to the real-time conversion of any text information into a standard fluent voice read out. It involves many disciplines and technologies such as acoustics, linguistics, digital signal processing, and computer science. The main problem to be solved is how to convert text information into audible sound information.

[0003] With the development of speech synthesis technology, users have higher and higher requirements for the diversification and differentiation of synthesized voice timbres. The existing method of generating new timbres is generally to obtain new timbres by customizing a new speaker's voice bank. However, the process of customizing a new speaker library is relatively complicated, and there are problems ...

Images