Asymmetrical voice conversion method based on deep neural network feature mapping

Abstract

The invention discloses an asymmetrical voice conversion method based on deep neural network feature mapping, and belongs to the technical field of voice conversion. The asymmetrical voice conversion method based on deep neural network feature mapping disclosed by the invention is specific to asymmetrical data of source voice and target voice. The method comprises the following steps: firstly, performing probability modeling by using the pre-training function of a deep network, and extracting high-order statistics features in a voice signal to provide a standby preferred space of network coefficients; secondly, performing incremental learning by using a small quantity of asymmetrical data, and correcting network weight coefficients by using an optimized transmission error to realize mapping of feature parameters. According to the asymmetrical voice conversion method, a network coefficient structure is optimized and is taken as a parameter initial value of a deep forward prediction network, network structure parameters are further transmitted reversely and optimized in the incremental learning process of a small quantity of asymmetrical data, so that mapping of the personal feature parameters of a speaker is realized.

Description

technical field [0001] The invention belongs to the technical field of speech conversion, and in particular relates to an asymmetric speech conversion method based on deep neural network feature mapping. Background technique [0002] Speech conversion technology, simply put, is to transform the voice of a speaker (called the source) by some means to make it sound like another speaker (called the target). Speech conversion is an interdisciplinary subject branch, and its content not only involves knowledge in the fields of phonetics, semantics, and psychoacoustics, but also covers all aspects of the field of speech signal processing, such as speech analysis and synthesis, speaker recognition, and speech coding. and enhancements etc. [0003] The ultimate goal of speech conversion is to provide instant speech services that can automatically and quickly adapt to any speaker. This system requires little or no user training and can function well for all users and various conditio...

AI Technical Summary

Problems solved by technology

[0006] The purpose of the present invention is to overcome the problems that the voice conversion system in the prior art not only strictly limits the way users can make sentences, but also requires a large amount of data for training, and the voice quality after conversion is not ideal at the same time, and provides a The asymmetric voice conversion method based on deep neural network feature mapping adopts the technical solution provided by the present invention. In the actual environment, the voice conversion system faces the problem of sharp deterioration of system performance under the condition of asymmetric data and data volume scarcity. The above-mentioned The two relatively independent links are integrated into a unified theoretical framework for research. At the same time, the deep neural network is used to conduct unsupervised training on the original data, and the high-order statistical feature information contained in it is extracted. Directive prediction training, and finally improve the generalization performance of the speech conversion system in the real environment

Images