A speech emotion enhancement method for hearing aid

Shulan Xia, Jilin Wang

College of Electrical Engineering, Yancheng Institute of Technology, Yancheng 224051, Jiangsu, PR China

In this paper, emotional perception of the hearing-impaired patients for hearing aid is investigated, and a speech enhancement algorithm is proposed, which is text-independent and requires less and non-parallel training data. In addition, the conversion of prosodic and spectral parameters is also studied. The Eigenvoice Gaussian mixture model (EV-GMM) is used to transform the F0s and spectral parameters, which is built using multiple pre-stored sources emotional and target neutral speech sentences. In the training and testing stages, the duration modification is utilized to improve the performance of EV-GMM training and converted output quality and an adaptive median filter is proposed to smooth the trajectory of the converted speech. Perceptual and objective experiments are presented, simulation results corroborate the effectiveness of the proposed algorithms.