Application of Environment Noise Classification towards Sound Recognition for Cochlear Implant Users

The ability of cochlear implant (CI) users in speech recognition decreases significantly in background noise. Various approaches have been proposed that optimize algorithms for specifically determined noisy environments with the ability to restore speech for cochlear implant users. This paper presents an approach to classifying different noise environments in our daily lives such as factory floor, jet cockpit, babble noise and etc. The noise classification system described here can be used to recognize different background noises and then optimize the coding strategies for hearing aids and cochlear implant devices. Seven types of noise environments are selected as the training data, and noise segments randomly cut from different noise recordings will be used as the test data. Classifiers based on Gaussian Mixture Models and Bayesian classifiers are developed and evaluated as well as KNN clustering. Features are extracted using MFCC feature extraction. In this paper, we aim to describe the automated solution for noise reduction in known types of noisy environments, and implement models in cochlear implant device. It is shown that training the classifier with 80% of the data resulted in 100% classification performance of all classes except the babble noise. By employing feature sub selection, the performance of the classifier was examined for every class using each single feature, and the role of each of the features in classifying each class was quantified. It was also found that by using only two of the features 100% performance could be enhanced for all classes except two of them.