Many-to-many speaker conversion method based on DenseNet STARGAN

Abstract

The invention discloses a many-to-many speaker conversion method based on DenseNet STARGAN, and the method employs the combination of the STARGAN and DenseNet to achieve a voice conversion system, andintroduces a GELU activation function into the STARGAN. On one hand, the DenseNet is used for solving the problem of network degradation in the training process, the back propagation of the gradientin the training process is facilitated; the deep network training efficiency is improved, on the other hand, a GELU activation function is used for replacing a ReLU activation function which is conventionally used, higher nonlinear representation capability is realized, the defect that the ReLU is in an inactive state in a negative interval is effectively overcome; the problem of network degradation in the training process is further relieved; according to the method, the representation capability of the STARGAN model is enhanced, the personality similarity and the voice quality of the converted voice are well improved, the high-quality many-to-many speaker voice conversion method is realized, and the method has a good application prospect in the fields of cross-language voice conversion,film dubbing, voice translation and the like.

Description

technical field [0001] The invention relates to a many-to-many speaker conversion method, in particular to a many-to-many speaker conversion method based on DenseNet STARGAN. Background technique [0002] Speech conversion is an important research branch in the field of speech signal processing, which is developed and extended on the basis of speech analysis, synthesis and speaker recognition. The goal of voice conversion is to change the voice personality of the source speaker so that it has the voice personality of the target speaker, while retaining semantic information, that is, to make the voice of the source speaker sound like the voice of the target speaker after conversion. . [0003] The initial stage of speech conversion is mainly speech conversion under parallel text. Parallel text requires that the source speaker and the target speaker need to utter sentences with the same speech content and speech duration, and the pronunciation rhythm and emotion are as consis...

AI Technical Summary

Problems solved by technology

[0007] Purpose of the invention: the technical problem to be solved by the present invention is to provide a multi-to-many speaker conversion method based on DenseNet STARGAN, which solves the network degradation problem in the training process of the existing method, accelerates the convergence speed of the model, and greatly improves the The representation ability of the model is enhanced, and at the same time, the difficulty of learning the semantic features of the encoding network is reduced, the learning function of the deep spectrum features of the model is realized, the spectrum generation quality of the decoding network is improved, and the semantic features and the individual characteristics of the speaker are fully learned. GELU The (Gaussian Error Linear Units) activation function is used as the activation function of the STARGAN model to help solve the problem of gradient disappearance in the training process of the deep network, improve the training efficiency of the deep network, and accelerate its convergence speed

Images