Chinese speech synthesis method and device, equipment and storage medium

Abstract

The invention relates to the field of language signal processing in the field of artificial intelligence, and discloses a Chinese speech synthesis method and devices, equipment and a storage medium. The training time is shortened, the model expressiveness and generalization ability are enhanced, and the synthesized speech quality is further improved. The method comprises the steps that an initialMeier spectrum and the a target vector are obtained; the target vector is processed to obtain a first sequence which is a two-dimensional tensor; the initial Meier spectrum is processed to obtain a target Meier spectrum; the target corresponding relationship between the first sequence and the target Meier spectrum in each subspace is determined; speech synthesis is conducted according to a self-attention mechanism and the target corresponding relationship, and the target speech is obtained.

Description

technical field [0001] The invention relates to the field of speech signal processing, in particular to a method, device, equipment and storage medium for synthesizing Chinese speech. Background technique [0002] At present, most speech synthesis research at home and abroad is aimed at text-to-speech conversion systems, and can only solve the problem of converting written language into spoken language output in a certain reading style, lacking the performance of different age, gender characteristics, tone, and speech speed, let alone endowed Personal emotional color. With the development of the needs of the information society, higher requirements are put forward for human-computer interaction, and the research of human-computer oral dialogue system is also mentioned on the agenda. [0003] Speech synthesis research has begun to develop from the conversion stage of text to speech to the conversion stage of concept to speech. This not only puts forward higher requirements ...

AI Technical Summary

Problems solved by technology

At present, for Chinese speech synthesis, when performing speech synthesis, the words and sentences are unnatural, the voice is dull, and the sense of rhythm is poor, which affects the quality of synthesis. There is a significant difference between the synthesized voice and the real human voice

[0004] Nowadays, long short term memory (LSTM) and other recurrent neural network (RNN) structures are commonly used in speech synthesis, which leads to the need to rely on the results of the previous time step during training, which is difficult to parallelize. too long

Images