Acoustic model post-processing method based on probability diffusion model, server and readable memory

Abstract

The invention discloses an acoustic model post-processing method based on a probability diffusion model, a server and a readable memory. The method comprises the following steps: model training: training the probability diffusion model by using the server, optimizing parameters of the probability diffusion model by reducing a loss function until the model converges, and obtaining the weight of the probability diffusion model; and model inference: according to the model weight obtained in the training stage, realizing spectrum optimization on the input predicted spectrum by using a server. According to the method, by learning the feature similarity between an input predicted spectrum and a real spectrum and using the data fitting capability of a noise estimation network in a model, probability distribution transfer based on diffusion is realized, and finally, the input predicted spectrum is more approximate to the real spectrum. And the naturalness of the synthesized speech is improved by improving the quality of the frequency spectrum. According to the method, the spectrum detail optimization effect can be achieved for the spectrums obtained by various acoustic models, and compared with other methods, the better spectrum generation effect is achieved.

Description

technical field [0001] The invention relates to the technical field of artificial intelligence. Aiming at the problem that the quality of speech synthesis needs to be improved, an acoustic model post-processing method based on a probability diffusion model is proposed. By optimizing the details of the predicted spectrum, the naturalness of the synthesized speech is improved. Background technique [0002] With the continuous development of artificial intelligence technology, problems in various fields based on artificial intelligence technology have made great progress. Speech synthesis is an important direction in artificial intelligence technology. Speech synthesis refers in detail to the synthesis from text to audio. The goal of speech synthesis technology is to enable machines to speak like people. High-quality speech synthesis technology can meet the needs of human-computer interaction in the current society. [0003] The existing speech synthesis technology first con...

AI Technical Summary

Problems solved by technology

However, there is a characteristic that the quality of the obtained acoustic spectrum is not high enough, so the synthesized speech wave file will also have a correspondingly low degree of naturalness

Images