Speech synthesis model generation method and device and electronic equipment

Abstract

Embodiments of the invention provide a speech synthesis model generation method and apparatus, and an electronic device. The method comprises the steps of obtaining a plurality of text samples; the text sample is trained, a first speech synthesis model is obtained, the first speech synthesis model comprises a plurality of convolution processing modules, and each convolution processing module comprises a plurality of parallel convolution layers; when i is an integer from 1 to n, determining a convolution layer equivalent to the plurality of parallel convolution layers included in the ith convolution processing module as an ith target convolution layer; replacing a plurality of parallel convolution layers included in an ith convolution processing module of the first speech synthesis model with an ith target convolution layer; and when a plurality of parallel convolution layers included in the nth convolution processing module are replaced by the nth target convolution layer, a second speech synthesis model is obtained. Therefore, the second speech synthesis model is obtained by equivalently transforming the target convolutional layer, so that the memory of the model in later application is saved.

Description

technical field [0001] The present application relates to the field of computer technology, in particular to a method, device and electronic equipment for generating a speech synthesis model. Background technique [0002] At present, the synthesis scheme of the model usually focuses on the combination of multi-way parallelism and attention structure in structure. The advantage of this combination design is that it can obtain faster convergence speed during training, and thanks to the information modeling of the entire training sample by the Attention structure, its synthesis effect is delicate and has a strong sense of anthropomorphism. [0003] For example, in the acoustic model synthesis scheme of speech synthesis technology, a large number of Attention and multi-channel parallel structures are used in the Fast Speech model, which will help optimize the gradient update of the model during the model training phase and avoid the disappearance of the gradient of the underlyin...

AI Technical Summary

Problems solved by technology

[0005] Embodiments of the present application provide a method, device, and electronic device for generating a speech synthesis model to solve the problem that the model combined with Attention and multi-channel parallel structure will occupy a large amount of memory in the later application

Images