Speech synthesis method, device and equipment and storage medium

Abstract

The invention discloses a speech synthesis method and device, equipment and a storage medium, and the method comprises the steps: obtaining an original text, a phoneme sequence corresponding to the original text, and the speaker features of to-be-synthesized speech, carrying out the feature fusion of the original text and the phoneme sequence, and obtaining a fusion feature; and performing encoding and decoding processing based on the fusion feature and the speaker features to obtain an acoustic spectrum, and performing speech synthesis based on the acoustic spectrum to obtain a synthesized speech. The fusion feature is obtained by fusing the original text and the phoneme sequence, input information is enriched, specific pronunciation information of different languages can be mined, for example, tone types of Chinese, tone nucleuses of Japanese, accent of Russian and the like can be displayed through the phoneme sequence, the acoustic spectrum is obtained, speech synthesis is carried out, the obtained synthesized speech is more natural, and accords with the pronunciation characteristics of the corresponding language, and the quality of the synthesized speech is higher.

Description

technical field [0001] The present application relates to the technical field of speech signal processing, and more particularly, to a speech synthesis method, apparatus, device and storage medium. Background technique [0002] In recent years, end-to-end speech synthesis systems have been able to achieve good results and can generate synthetic speech close to human in real time. With the development of globalization, in important scenarios of speech synthesis applications such as social media, informal information, and voice navigation, the language phenomenon of mixing different languages ​​in text or speech becomes more and more obvious. One wants to synthesize these cross-lingual sentences with a consistent and natural voice, but most current end-to-end models assume that the input is monolingual and use only raw text as the input to the synthesis model. [0003] The inventor of the present case found that different languages ​​have different pronunciation phenomena, su...

AI Technical Summary

Problems solved by technology

People hope to synthesize these cross-lingual sentences with a consistent and natural voice, but most current end-to-end models assume that the input is a single language and only use the original text as input to the synthesis model

[0003] The inventors of this case found that the pronunciation phenomena of different languages ​​are different, such as Chinese tone patterns, Japanese accents, and Russian accents, etc. are not expressed on the text, so the existing synthesis model for a single language only uses the original Text is used as model input, and for speech synthesis of cross-lingual sentences, the quality of the synthesized speech is not high

Images