Robust speech / music classification in audio documents

This paper deals with a novel approach to speech / music segmentation based on four original features: 4 Hz mod-ulation energy, modulation entropy and two segmental pa-rameters. The relevance of these features is evaluated in a first experiment based on a development corpus composed of collected samples of speech and music. Another corpus is employed to verify the robustness of the employed algo-rithm. This experience is made on a TV movie soundtrack and shows performances reaching a correct identification rate of 90 %. 1. MOTIVATIONS To describe and index an audio document, key words or melodies are semi-automatically extracted and speakers are detected. More recently, the problem of topics retrieval ha