Exploring features for audio clip classification using LP residual and AANN models

In this paper, we demonstrate the presence of audio-specific information in linear prediction (LP) residual, obtained after removing the predictable part of the signal. It is known that the residual of a signal is less subject to channel degradations as compared to spectral information. So systems built using the residual may be robust against degradations. This emphasizes the importance of information present in the LP residual of audio signals. But it is difficult to extract information from the residual using known signal processing algorithms. Autoassociative neural networks (AANN) models have been used to capture the distribution of feature vectors for pattern recognition tasks. In this paper, AANN models have been shown to capture audio-specific information from the LP residual of signals to classify audio data.