Improved Sound Classification by Means of Sound Localization in Hearing Devices

Abstract

Hearing devices are becoming a more and more common solution for people with mild to profound hearing loss. The hearing instruments are, indeed, able to compensate for the individual hearing loss while trying to provide the best hearing experience possible. However, the acoustic environments which the average user experiences during the day are usually very different and complex. Most of the devices try to cope with these big variations in the surroundings by adjusting the settings based on some parameters senses by the microphones. One of the most common solutions, in this regards, is the sound classifier. The main objective of this tool is to blindly understand the acoustic scenario around the hearing aid user and enable the appropriate change in the settings based on the detected environment. Although the sound classifier is a precious tool in a hearing device, very often it does not consider a very important aspect of sound, i.e., sound is generated by sources placed in a 3D world. The spatial components in the acoustic scenarios are often discarded after being sensed by the microphones. In this work, a possible solution to integrate spatial components to the sound classifier in order to have a better understanding of the acoustic environment was investigated. The solution here proposed is to extend the inputs of the sound classifier with additional informations about the location of the sound sources present in the environment. To provide these additional informations, a localization algorithm tailored for hearing devices applications has been developed. The algorithm creation was inspired by the human auditory system which relies on the estimation of the binaural cues. In addition to the localization algorithm, a blind direct-to-reverberant energy ratio for speech signal has been developed. Moreover, an exploratory solution for head tracking based on a 3-axis accelerometer was investigated.