Robust Speaker Recognition in Unknown Noisy Conditions

This paper investigates the problem of speaker identification and verification in noisy conditions, assuming that speech signals are corrupted by environmental noise but knowledge about the noise characteristics is not available. This research is motivated in part by the potential application of speaker recognition technologies on handheld devices or the Internet. While the technologies promise an additional biometric layer of security to protect the user, the practical implementation of such systems faces many challenges. One of these is environmental noise. Due to the mobile nature of such systems, the noise sources can be highly time-varying and potentially unknown. This raises the requirement for noise robustness in the absence of information of the noise. This paper describes a method, named universal compensation (UC), that combines multi-condition training and the missing-feature method to model noises with unknown temporal-spectral characteristics. Multi-condition training is conducted using simulated noisy data with limited noise varieties, providing a “coarse” compensation for the noise, and the missing-feature method refines the compensation by ignoring noise variations outside the given training conditions, thereby reducing the training and testing mismatch. This paper is focused on several issues relating to the implementation of the UC model for real-world applications. These include the generation of multi-condition training data to model real-world noisy speech, the combination of different training data to optimize the recognition performance, and the reduction of the model’s complexity. Two databases were used to test the UC algorithm. The first is a re-development of the TIMIT database by re-recording the data in the presence of various noises, used to test the model for speaker identification with a focus on the noise varieties. The second is a handheld-device database collected in realistic noisy conditions, used to further validate the model on the real-world data for speaker verification. The new model was compared to baseline systems and has shown improved identification and verification performance. J. Ming is with the School of Electrical Engineering and Computer Science, Queen’s University Belfast, Belfast BT7 1NN, U.K. (e-mail: [email protected], phone: 44-28-90974723; fax: 44-28-90975666). T. J. Hazen and J. R. Glass are with the MIT Computer Science and Artificial Intelligence Laboratory, Cambridge, MA 02139, U.S.A. (e-mail: hazen/[email protected]; phone: 1-617-253-4672/1640; fax: 1-617-258-8642). D. A. Reynolds is with the MIT Lincoln Laboratory, Lexington, MA 02420, U.S.A. (email: [email protected]; phone: 1-781981-4494; fax: 1-781-981-0186). The work was sponsored in part by Intel Corporation, and in part by the Department of Defense under Air Force Contract FA8721-05-C-0002. Opinions, interpretations, conclusions, and recommendations are those of the authors and are not necessarily endorsed by the United States Government. November 10, 2005 DRAFT