Extraction, Identification and Regeneration of Source Audio Signal Using HHT and SVM

This paper presents, the working idea of Separation of each Individual Components that make up a Mixed Audio Signals, Identification of each of the Extracted Components and finally Wave Regeneration of the components that was separated. That is if the Mixed audio Signal contains components such as "Guitar" and "Violin". Then each of the Components of Guitar and Violin will be separated and the individual components will be identified. The Development of this particular working was done using Mat lab Simulation and the results were quite fruitful. The Extraction of each of the components can be carried out by Empirical mode Decomposition (it is a part of Hilbert Hyuang Transform) and the Separation to its constituent clusters or classes can be done by Fuzzy C-means Clustering Technique. The Separated components were identified by the SVM Classification technique. The SVM classifier's Training and Testing Sets were modified in such way, so that it can predict sound data. The Prediction output of the SVM Classification is the class to which the Extracted Component belongs (i.e. Whether the extracted components are guitar or Flute or Violin etc.). The Extracted Components are smoothed using B-spline Curve Fitting for the Wave regeneration.