Speech synthesis method and device

Abstract

The embodiment of the invention provides a speech synthesis method and device, and the method comprises the steps: obtaining an input text and a style generation module, and obtaining a style feature representation according to the input text and the style generation module; obtaining a tone mark and a target speech synthesis model; wherein the target speech synthesis model is obtained by training based on a pre-trained duration extraction model used for extracting related duration representation and a style feature module used for extracting style feature representation; and according to the input text, the style feature representation, the tone mark and the speech synthesis model, obtaining an emotional speech with a corresponding style and a corresponding tone. The emotion speech synthesis of the custom style is realized through the extraction and preprocessing mode of the custom emotion feature representation, the effect that the style of the synthesized speech is suitable for all timbres in a corpus is realized on the basis of supporting the custom of the synthesized speech style, and the purpose of the semi-supervised emotion speech synthesis of the custom style is achieved.

Description

technical field [0001] The invention relates to the technical field of information processing, in particular to a speech synthesis method and a speech synthesis device. Background technique [0002] In recent years, with the rapid progress in machine learning, especially related technology research in the field of deep learning, it has promoted a great change in the way of human-computer interaction, and it is also accompanied by the landing of more and more commercial products. As a new model, voice interaction not only brings a new user experience, but also expands the design ideas and application scenarios of various products. The speech synthesis system is one of the core systems of the closed-loop speech interaction. How to make the synthesized speech contain richer emotions is an important factor for personalized speech synthesis and improving the expressiveness of speech synthesis. [0003] The existing emotional speech synthesis method can realize emotional speech s...

AI Technical Summary

Problems solved by technology

[0003] The existing emotional speech synthesis method can realize emotional speech synthesis of a unified style by recording and making a corpus of a unified style, and can realize a variety of emotional speech synthesis of a fixed style by recording and making a corpus of a variety of fixed styles. Make a variety of small corpora with fixed styles, and perform speech style conversion or speech emotion conversion on other corpora with a single style. Existing Style Limits

Images