A New Approach for Automatic Audio Segmentation , And Reconstruction

Automatic Audio Segmentation aims at extracting information about type of audio i.e. silence, clean speech or speech with noise, music etc. The aim of this thesis is to find and extract different features of audio, segment the audio and combine them to form single type of audio. In this work, the audio signal is initially decomposed into non-overlapping frames. Then these frames are decomposed into a set of band-limited functions known as Intrinsic Mode Functions (IMF) using Empirical Mode Decomposition (EMD). Temporal and Spectral features then extracted from these IMFs and thereafter classification is done using k-nearest neighbour (kNN) classifier. Depending upon classification accuracy of the features the segmentation is done using k-means clustering algorithm. Then the segments were combined to form new audio signal. KEYWORDS--Intrinsic Mode Function, Empirical Mode Decomposition, Spectral And Temporal Features, k-NN classifier, K Means Clustering.