CBFMCycleGAN-VC: Using the Improved MaskCycleGAN-VC to Effectively Predict a Person’s Voice After Aging

One task of nonparallel speech conversion is to convert the source speaker’s samples target samples, keeping content unchanged. In view advantages MaskCycleGAN-VC in conversion, such as small model size and superior performance, our paper uses basic structure improve it proposes a cyclic boundary method filling frame (CBFMCycleGAN-VC) model, which predicts voice person he ages by using his younger self. First, this adds preprocessing modules, including Chebyshev low-pass filter adaptive filter, increases robustness system. Second, considers time-domain difference weight parameters, makes easier grasp mapping law structure, with faster convergence speed. Last, circular introduced avoid ringing effect, enhance connection between filled adjacent frame, obtain better generator. The simulation results show that CBFMCycleGAN-VC more suitable for predicting voices elderly people, speed faster. converted also closer speaker time domain frequency domain. Under condition accuracy rate similar MaskCycleGAN-VC, MOS score 17.5% higher than MaskCycleGAN-VC.