Class-specific model mixtures for the classification of time-series

We present a new classifier for acoustic time-series that involves a mixture of generative models. Each model operates on a feature stream extracted from the time-series using overlapped Hanning-weighted segments and has a probability density function (PDF) modeled with a hidden Markov model (HMM). The models use a variety of segmentation sizes and feature extraction methods, yet can be combined at a higher level using a mixture PDF thanks to the PDF projection theorem (PPT) that converts the feature PDF to raw time-series PDFs. The effectiveness of the method is shown using an open data set of short-duration acoustic signals.