Fig. 4From: A multi-task learning speech synthesis optimization method based on CWT: a case study of Tacotron2Wavelet basis function fitting non-stationary signals through translation and stretchingBack to article page