Speech Enhancement Using Generative Adversarial Network (GAN)

Abstract

AbstractMost of the restoration techniques for loss of voice result in whispered and monotonous speech. In addition to intelligibility, this type of speech is poor in expressiveness and naturalness due to a) the lack of pitch resulting in whispered speech, and b) artificial pitch production resulting in monotone speech. This research work offers a neural network method for estimating a fully voiced speech waveform from alaryngeal whispering speech waveform. In this research paper a speech enhancement method using Generative Adversarial Networks (GANs) is implemented. The aim of this GAN implementation to perform whispered-to-voiced speech conversion and to handle speech reconstruction tasks.KeywordsNeural impairmentDeep learningSpeech processingDigital health