Speech synthesis method and device

Abstract

The invention discloses a speech synthesis method, and belongs to the technical field of speech synthesis. The method comprises the following steps: performing rhythm analysis on a to-be-synthesized text; segmenting a long text into a short text set with a proper length according to a rhythm analysis result, and recording a text segmenting sequence; calling a voice synthesis model for the text objects in the text set to generate acoustic features in parallel; splicing the acoustic features obtained by the text objects according to the text segmenting sequence; and enabling the spliced completeacoustic features to pass through a vocoder model, and finally, outputting an audio. On the basis of traditional speech synthesis methods, parallel processing in the speech spectrum generation process of a to-be-synthesized text is reasonably utilized, the problems that when the text is too long, the speech synthesis speed is low, and speech synthesis model spectrum synthesis is prone to failuredue to the fact that the text is too long are effectively solved, the speech synthesis speed is effectively increased, so that the speech synthesis system is more efficient, stable and natural.

Description

technical field [0001] The invention relates to the technical field of speech synthesis, in particular to a speech synthesis method and device. Background technique [0002] Speech synthesis is a technology that converts text information into speech information. Its process mainly includes: processing text, such as text preprocessing, word segmentation, prosody prediction, phoneme labeling, etc., and then training the acoustic model, using Mel spectrum or linear The spectrum is used as an acoustic feature, and finally a vocoder is used to synthesize the spectrum into sound. [0003] At present, the end-to-end (End-to-End) modeling method represented by Tacotron has become the mainstream. Tacotron is a mid-section structure proposed by Google that combines the original duration model and acoustic model, and can connect any TTS front-end and back-end. The TTS front-end includes text regularization, prosody prediction, phoneme conversion, etc., while the TTS back-end mainly re...

AI Technical Summary

Problems solved by technology

[0006] The purpose of the present invention is to provide a method and device for speech synthesis to improve the stability and efficiency of the speech synthesis system, in order to solve the technical problems in the background art: when the text is too long in the speech synthesis process, the speech synthesis speed is slow, and due to An issue where the text was too long and the audio could not be synthesized accurately

Images