VAE (Variational Autoencoder)-based voice conversion method under non-parallel corpus training

Abstract

The invention discloses a VAE (Variational Autoencoder)-based voice conversion method under a non-parallel corpus training condition. Under a non-parallel text condition, Bottleneck features are extracted through a deep neural network, and then, based on a VAE model, learning and modeling of a conversion function are realized, and in a conversion stage, conversion from multiple speakers to multiple speakers can be realized. The VAE-based voice conversion method under non-parallel corpus training has the advantages in three aspects: 1) dependence on a parallel text is relieved, and any alignment operation is not needed during a training process; 2) multiple source-target speaker pair conversion systems can be integrated to one conversion model, and multi-to-multi conversion is realized; and3) the multi-to-multi conversion system under the non-parallel text condition can provide a techncial support for enabling the technology to go to actual voice interaction.

Description

technical field [0001] The invention belongs to the field of speech signal processing, and in particular relates to a speech conversion method based on a variational autoencoder (VAE) model under non-parallel corpus training. Background technique [0002] Speech conversion technology is a research branch of speech signal processing. It covers the fields of speaker recognition, speech recognition and speech synthesis. It intends to change the personalized information of speech while keeping the original semantic information unchanged, so that The speech of a particular speaker (ie, the source speaker) sounds like the speech of another particular speaker (ie, the target speaker). The main tasks of speech conversion include extracting the characteristic parameters of two specific speaker's voices and performing mapping transformation, and then decoding and reconstructing the transformed parameters into converted speech. In this process, it is necessary to ensure the accuracy o...

AI Technical Summary

Problems solved by technology

However, there are still some defects in this type of algorithm. For example, the classic method of using Gaussian mixture model for speech conversion is mostly based on one-to-one speech conversion tasks, requiring the source speaker and the target speaker to use the same training sentence content. Only by aligning the spectral features with Dynamic Time Warping (DTW) frame by frame can the mapping relationship between spectral features be obtained through model training. Such a voice conversion method is not flexible enough in practical applications; the Gaussian mixture model is used to train the mapping function When considering global variables and iterating the training data, the calculation amount will increase sharply, and only when the training data is sufficient, the Gaussian mixture model can achieve a better conversion effect, which is not suitable for limited computing resources and equipment

Images