Sound excerpt based similarity search in P2P applications

Abstract

The increasing popularity of file sharing systems was practically grounded by music recordings becoming widely available. The successful introduction was closely followed by new systems, all utilizing search methods essentially based on meta-data. In this article a measure for audio similarity is proposed. The frequency of the amplitude values of the sound wave signal is determined and utilized in a previously developed content searching procedure to achieve automatic classification. The new, low complexity inspection method enables to determine more local characteristics then the amplitude-histogram, such as the silence periods. The length of the determined periods and their relative position increase the dimension of the similarity measure.