Speech Activity Detection Implementation in Hearing Aids Using Deep Belief Network

Abstract

Hearing aids are small devices, hearing impaired and deaf people wear to compensate their hearing losses. Most of the affordable available hearing aids nowadays are designed only on one environment and for certain hearing characteristics. In our work, we proposed an intelligent hearing aid that is able to adjust to changing environment and it can be designed according to the patient hearing characteristic. Speech activity detection is implemented using deep belief networks. This step is used so that different processing steps will be done to the audio signal before it reaches the human ears. Two different features have been used to train the DBN and a comparison was made between them. MFCC and LPCC features were used and gave similar output results. A choice of using LPCC features is made because of they are easy to implement.