Autoassociator-Based Models for Speaker Verification
M. Gori, L. Lastrucci, and G. Soda
Dipartimento di Sistemi e Informatica, Universit?a di Firenze
Via di Santa Marta 3 - 50139 Firenze - Italy
Tel. +39 (55) 479.6265 - Fax +39 (55) 479.6363
E-mail : (marco,luca,giovanni)@mcculloch.ing.unifi.it
In this paper, we propose an autoassociator-based connectionist model that turns out to be very useful for problems of pattern verification. The model is based on feedforward networks acting as autoassociators trained to reproduce patterns presented at the input to the output layer. The verification is established on the basis of the distance between the input and the output vectors. We give experimental results for assessing the effectiveness of the model for problems of speech verification. The performances were evaluated on DARPA-TIMIT database in continuous speech, using different thresholds and preprocessing schemes, with very promising results.
Index Terms- Connectionist models, continuous speech, nonlinear autoassociators, speaker verification,
One fundamental problem in pattern recognition is that of verifying the authenticity of a given pattern. Examples of similar problems can easily be found in the real world, where machines are increasingly employed for controlling the access of people on the basis of voice, faces, and fingerprints. Other significant examples involve non-human patterns that, like banknotes, must not simply be recognized, but overall, verified against imitations.
In this paper, we propose a connectionist model for verifying the authenticity of patterns and, in particular, we suggest its usage for open set speaker recognition problems . In that case, one wants to decide whether the speakers of a test utterance belong to a group of N known speakers. Unlike the closed set problem, the machine must be protected against any potential impostor, whose identity is not known in advance. Speaker verification is essentially an open set problem where additional checks,