Hearing environment recognition in hearing aids

The hearing environment recognition algorithm is one of the core algorithms of the modern digital hearing aids. It can recognize the quiet, noisy and musical environment under a variety of scenarios, such as street, train or restaurant scene in noise environment. It then adjusts the parameters of other hearing aids algorithms according to different audio environment. This paper presents a method to improve the training algorithm of Gaussian mixture models(GMMs), which uses fusion LBG clustering algorithm and annealing algorithm to initialize the value of GMM parameter means μ, instead of initializing it randomly in traditional way. The method can improve recognition accuracy by optimizing the GMM parameters. The experimental result demonstrates that the accuracy rate of the traditional GMMs algorithm is about 77%, while that of the method we proposed is about 85%. In the end, we continue to explore the influence of the selection of the MFCC features and Gaussian mixture numbers on the recognition rate.