Modeling and classification of natural sounds by product codehidden Markov models |
| |
Authors: | Woodard J.P. |
| |
Affiliation: | Autonetics, Anaheim, CA; |
| |
Abstract: | Linear predictive coding (LPC), vector quantization (VQ), and hidden Markov models (HMMs) are three popular techniques from speech recognition which are applied in modeling and classifying nonspeech natural sounds. A new structure called the product code HMM uses two independent HMM per class, one for spectral shape and one for gain. Classification decisions are made by scoring shape and gain index sequences from a product code VQ. In a series of classification experiments, the product code structure outperformed the conventional structure, with an accuracy of over 96% for three classes |
| |
Keywords: | |
|
|