Hierarchical organization of a set of Gaussian mixture speaker models for scaling up indexing and retrieval in audio documents

  • Authors:

  • J. E. Rougui;M. Rziza;D. Aboutajdine;M. Gelgon;J. Martinez

  • Affiliations:

  • Mohammed V University, RP Rabat- Morocco;Mohammed V University, RP Rabat- Morocco;Mohammed V University, RP Rabat- Morocco;l'université de Nantes, Nantes cedex, France;l'université de Nantes, Nantes cedex, France

  • Venue:
  • Proceedings of the 2006 ACM symposium on Applied computing
  • Year:
  • 2006

Quantified Score

Hi-index 0.00

Visualization

Abstract

This work considers the case of spoken radio archives, in which a continuous ow of speech is introduced into the indexing system. To retrieval segments originating from a given speaker, scanning the set of enrolled speaker is needed. We propose a technique for organizing hierarchically the set of speaker models, to reduce the cost of speaker identification. This organization can be obtained and updated at low cost, thanks to a particular algorithm that exploits the Gaussian mixture form of each speaker model. This parcimony is due to having computations carried out of model parameters rather than feature vectors. A first validation on real data illustrates the proposal.