Audio Denoising, Recognition and Retrieval by Using Feature Vectors

Abstract

In this paper, we present the study of audio signal denoising, recognizing its audio type and retrieving similar signals from the database. Primarily the signal is denoised to remove any noise present in the unknown signal. Time Frequency Block Thresholding is used to achieve this. Removing the noise from audio signals requires processing of time-frequency coefficients in order to keep away from producing the musical noise. A block thresholding estimation procedure is introduced, which relates the parameters adaptively to signal property. Content Based Audio Retrieval system is very useful to identify the unknown audio signals. Audio signals are classified into music, speech and background sounds. This is done by the use of various feature vectors such as zero crossing rate, spectral centroid, spectral flux, spectral roll off. Use of multiple feature vectors improves the accuracy of the results being retrieved from the audio database. Various experiments demonstrate the performance and accurateness, providing good results for non standard signals.