Reconstruction of Normal Speech from Whispered Speech Based on RBF Neural Network

Restriction of normal speech from Chinese whispered speech based on radial basis function neural network (RBF NN) is proposed in this paper. Firstly, capture the nonlinear mapping of spectral envelope between whispered and normal speech by RBF NN; secondly, modify the spectral envelope of the whispered speech by adopting the trained neural network; finally, convert the whispered speech into normal speech by using the linear spectral pairs (LSP) synthesizer. Both subjective and objective assessments are conducted on the converted speech quality. Simulation results show that the score of the Mean Opinion Score (MOS) is 3.2; the distorted distance of bark spectrum is decreased. Both intelligibility and quality of the converted speech are satisfied.