Audio indexing: primary components retrieval

This work addresses the soundtrack indexing of multimedia documents. We present and merge two audio classification tools that we have developed. The first one, a speech music classification tool, is based on three original features: entropy modulation, stationary segment duration and number of segments. They are merged with the classical 4 Hertz modulation energy. The other one, a jingle identification tool, used an Euclidean distance in a spectral domain to structure the audio data flow. The system is tested on TV and radio corpora. Systems are simple, robust and could be improve on every corpus without training or adaptation.

• Type: Journal
• Year: 2006
• Where: MTA
• Authors: Julien Pinquier, Régine André-Obrecht
• Keywords: Audio Classification Tools, Classification Tool, Hertz Modulation Energy, MTA 2006