A Review Of Deep Learning Applications For Speech Processing Improvement

Abstract

Improve the quality of the spoken word is a common goal for many audio and speech signal processing applications. A noisy voice signal's quality and understandability may be improved via speech augmentation. Speech augmentation is critical in a wide range of fields, including hearing aids, ASR, and mobile communication. DNN-based architectures for speech recognition and augmentation have shown to be quite effective in recent years, according to a new study. In the actual world, where many disturbances may concurrently contaminate speech, we examine the issue of speech improvement in this study. Current research on speech improvement focuses mostly on the existence of single noise in damaged speech, which is far off from the real-world situations that really exist. In particular, we are concerned with enhancing the clarity of workplace speech in which several stationary and non-stationary sounds may be present at the same time. In some circumstances, Deep Neural Networks (DNN) may be utilised to enhance speech. For the improvement of loud speech, we also look at DNN training using psychoacoustic models from speech coding.