Unsupervised environmental sound recognition

摘要

Environmental sound recognition is an audio scene identification process to locate a person by analyzing the background sound. This paper deals with the prototype modeling of environmental sound recognition that is based on unsupervised learning. The unsupervised learning finds a hidden structure in a group of data given as input. There is no need of a label to which the input data belongs. So this could be used for the practical cases. Sound recognition involves the collection of audio data, extraction of significant features and finding a common structure between them, thus leading to grouping of the data. The Mel frequency cepstrum coefficients are extracted. These features are used for clustering by a Gaussian mixture model which is a probabilistic model. The clustering leads to the identification of the correct audio scene. The implementation is done with the help of MATLAB and ModelSim. Five major environmental sounds which include the sound of car, office, restaurant, street, subway are considered. The parameters of the Gaussian mixture model are estimated in the training phase. The model is tested with the inputs considering the parameters. The MATLAB implementation shows an efficiency of 98%. The hardware implementation of the same shows an efficiency of 96.4%.