Speech synthesis model training method, speech synthesis method and device

Abstract

The embodiment of the present invention provides a speech synthesis model training method, speech synthesis method and device, the speech synthesis model training method includes: obtaining the first word vector sequence corresponding to the Chinese character sentence of encoding training; using the encoding module to encode the first word vector The sequence is encoded to obtain the first linguistic coding feature; the linguistic feature decoding module is used to decode the linguistic feature of the first linguistic coding feature to obtain the linguistic decoding feature; according to the relationship between the linguistic decoding feature and the reference linguistic decoding feature The linguistic feature loss of the speech synthesis model is adjusted by adjusting the model parameters of the coding module of the speech synthesis model until the linguistic feature loss meets the linguistic feature loss threshold, and the trained coding module of the speech synthesis model is obtained. The speech synthesis model training method, speech synthesis method and related devices provided by the embodiments of the present invention can reduce the complexity of speech synthesis, improve the training accuracy of the encoder, and ensure the effect of synthesized speech.

Description

technical field [0001] The embodiments of the present invention relate to the field of computers, and in particular to a speech synthesis model training method, device, device and storage medium, and a speech synthesis method, device, device and storage medium. Background technique [0002] With the development of artificial intelligence technology, people pay more and more attention to speech synthesis technology. Using speech synthesis technology and speech recognition technology, on the one hand, computers and other equipment can generate spoken language that people can understand; Human-machine voice communication can be realized. [0003] In order to realize speech synthesis, the traditional parametric speech synthesis method can be used, which is divided into multiple parts such as linguistic feature prediction, duration prediction, and acoustic feature prediction. It is necessary to build and train models for each part to realize speech synthesis. [0004] It can be ...

AI Technical Summary

Problems solved by technology

[0004] It can be seen that the above method requires the design and training of multiple models, so the training process is very complicated, and the effect of the synthesized speech obtained is not ideal, and the sound quality is flat

Images