IEEE AASP Challenge on Detection and Classification of Acoustic Scenes and Events IEEE AASP SCENE CLASSIFICATION CHALLENGE USING HIDDEN MARKOV MODELS AND FRAME BASED CLASSIFICATION

The IEEE AASP Challenge involves the detection and classification of acoustic scenes and events. The scene classification (SC) challenge consists of 10 different scenes of 10 audio files or length 30 seconds each, totaling a number of 100 audio clips. The list of scenes is: busy street, quiet street, park, open-air market, bus, subway-train, restaurant, shop/supermarket, office, and subway station. The goal is to test on a development set that is composed of audio clips of the same scenes as the training set and determine what scene the audio clips originated from. One of the algorithms presented in this paper to discriminate between these different scenes include the use of hidden Markov models (HMMs) and Gaussian mixture models (GMMs). The features that were used include the following: short time Fourier transform, loudness, and spectral sparsity. Using these features yielded 72% correct classification with 10 fold crossvalidation. The other algorithm implemented uses the same features as before plus temporal sparsity to classify individual frames of an audio clip, then vote on the class. This algorithm achieved 62% accuracy.