Sound classification based on modified log energy for digital hearing aids

An sound classification algorithm based on minimum distance clustering using modified log energy as features is presented. This system attempts to distinguish between three listening environment categories: traffic noise, babble noise, and quiet environment. Each sound source is assigned one codebook area beforehand. The decision-making is based on the distribution ratios of feature vectors falling into each codebook area. The feature group used in this system is compared to another two feature groups, a common cepstral feature group and mel-frequency cepstral coefficients (MFCCs), with the same classification method. The classifier using the modified log energy is superior to those using the other two feature groups in overall hit rate and the mean squared error between the classifier output and the correct output. Compared to MFCCs, the modified log energy set reduces computational burden of the whole system.