Segmentation de signaux audio: Une nouvelle approche utilisant le critere d'alignement

In audio indexing systems it's always needed to access directly to the particular acoustical event like musical record or a speaker excerpt, then we must to a priori design a binary based audio algorithm which permit to segregates the acoustic classes. This paper addresses a new method which combine the classical Principal Component Analysis (PCA) with the Alignment criterion introduced and often used in machine learning problems. This new method is model free and easy computed, we demonstrate its achievement and show their promising results which in return permits their use on DSP and FPGA platforms.