A non-parallel many-to-many speech conversion method based on STARGAN model

Speech conversion is a task of converting speaker personality features while maintaining semantic information so that the converted speech has the same semantic information as the source speech and has the same speaker personality features as the target speech. Traditional speech conversion methods require parallel speech data to train the conversion model, but it is difficult to obtain parallel data in practical scenarios. To break this limitation, many non-parallel speech conversion methods have been proposed. In this paper, we propose a speech conversion method based on STARGAN model, which can achieve many-to-many speech conversion under non-parallel conditions, and the quality and similarity of the converted speech are better.