Graph theory for the discovery of non-parametric audio objects

Abstract

A novel framework based on graph theory for structure discovery is applied to audio to find new types of audio objects which enable the compression of an input signal. It converts the sparse time-frequency representation of an audio signal into a graph by representing each data point as a vertex and the relationship between two vertices as an edge. Each edge is labelled based on a clustering algorithm which preserves a quality guarantee on the clusters. Frequent subgraphs are then extracted from this graph, via a mining algorithm, and recorded as objects. Tests performed using a corpus of audio excerpts show that the framework discovers new types of audio objects which yield an average compression gain of 23.53% while maintaining high audio quality.