Classification of Sound Signals via Computationally Efficient Supervised and Unsupervised Learning Schemes

Classification of sound signals is increasingly being used in hearing improvement devices such as hearing aids, cochlear implants, and smart headphones. Classification of sound signals enables adapting the speech enhancement/noise reduction algorithms in such devices to different sound environments in an automatic manner. The thrust of this dissertation research has been on the development of sound signal classification approaches that are computationally efficient, thus enabling their real-time deployment in hearing improvement devices. Both supervised and unsupervised learning schemes have been examined. For the supervised case, effective and computationally efficient features and classifiers have been developed. For the unsupervised case, an online clustering algorithm has been developed without knowing the number of clusters. Experimental results obtained indicate that the developed classification approaches outperform the existing sound classification approaches in terms of both classification rates and computational efficiency.