Fig. 3From: A multi-task learning speech synthesis optimization method based on CWT: a case study of Tacotron2Fourier transform fitting non-stationary signalsBack to article page