M.Sc. Tezi Görüntüleme

      The numerical audio data obtained from the music volume involve very complex structures and various styles. The purpose of this paper, is both to separate the audio data into segments based on the content of them using fuzzy clustering method and also to predict the region of each note. In the literature, there are many methods developed for segmentation. Some of these methods are: Mel Frequency Campestral Coefficients, ZeroCrossing, Average Magnitude Difference Function, Energy Threshold Value, Entropy and Discrete Fourier transform methods. The most commonly used method is segmentation process using threshold value. This method cannot segment efficiently in some situations provided that the amplitude of values create an increasing function or musical notes and start to play in succession or any made musical notes fade out. This method cannot desire segmentation efficiency. In this paper, a new method is proposed that uses segmentation. In this method, various feature coefficient extraction is performed by taking small frameworks on the audio data. The regions of the segment are determined using fuzzy clustering according to these feature coefficients, therefore, the silence regions are determined. In this case, silence zones join silence clusters taking low membership values. Therefore, each silence region becomes separated from the segment with a fuzzy cluster.