Person verification by fusion of prosodic, voice spectral and facial parameters

Hernando, Javier; Farrús, Mireia; Ejarque, Pascual; Garde, Ainara; Luque, Jordi

doi:http://dx.doi.org/10.5220/0002105200170023

Person verification by fusion of prosodic, voice spectral and facial parameters

Citation

Hernando J, Farrús M, Ejarque P, Garde A, Luque J. Person verification by fusion of prosodic, voice spectral and facial parameters. In: Malek M, Fernández-Medina E, Hernando J, editors. SECRYPT 2006, Proceedings of the International Conference on Security and Cryptography; 2006 Aug 7-10; Setúbal, Portugal. [Setubal, Portugal]: INSTICC Press; 2006. p. 17-23. DOI: 10.5220/0002105200170023

Abstract

Prosodic information can be used successfully for automatic speaker recognition, although most of the speaker recognition systems use only short-term spectral features as voice information. In this work, prosody information is added to a multimodal system based on face and voice characteristics in order to improve the performance of the system. Fusion is carried out by using various fusion strategies and two different fusion techniques: support vector machines and matcher weighting. Results are clearly improved when a previous normalization based on histogram equalization is done before the fusion of the monomodal scores.

Description

Comunicació presentada a: SECRYPT 2006, International Conference on Security and Cryptography, celebrada a Setúbal, Portugal, del 7 al 10 d'agost de 2006