Probabilistic neural networks combined with GMMs for speaker recognition over telephone channels

T. Ganchev1, A. Tsopanoglou2, N. Fakotakis1, G. Kokkinakis1
1Wire Connnunication Laboratory, University of Patras, Patra-Rio, Greece
2Knowledge S.A. (LogicDIS GROUP), Patras, Greece

Tóm tắt

We study the applicability of probabilistic neural networks (PNNs) as core classifiers to medium scale speaker recognition over fixed telephone networks. In particular, banking applications with up to 400 enrolled speakers and short training times are targeted. Two PNN-based open-set text-independent systems, for speaker identification and speaker verification, respectively, are presented. The performance of these systems is studied with and without the use of a supporting Gaussian mixture models classifier. Results from experiments carried out on the Polycost and SpeechDat(II)-Greek corpus, with training times as short as 43 seconds, are reported.

Từ khóa

#Neural networks #Speaker recognition #Telephony #Mel frequency cepstral coefficient #Speech #Training data #Working environment noise #Laboratories #Banking #Wideband

Tài liệu tham khảo

10.2307/2346830 10.1109/MASSP.1984.1162229 chatzi, 1997, Greek speech database for creation of voice driven teleservice, Proc Eurospeech 97, 4, 1755 petrovska, 2000, Poly-cost: A Telephone-Speech Database for Speaker Recognition, Speech Communication, 31, 265, 10.1016/S0167-6393(99)00082-5 hennebert, 1996, The POLYCOST 250 Database (v 1.0), COST25 0 report 10.1142/S0218001497000196 0, A list of known bugs in version 1.0 of POLYCOST data-base rabine, 1976, A Comparative Performance Study of Several Pitch Detection Algorithms, IEEE Transactions on ASSP, assp 24 georgila, 2000, SpeechDat Greek Database for the Fixed Telephone Network, Documentation included in the database CD-ROMs vivaracho, 2001, A Comparative Study of MLP-based Artificial Neural Networks in Text-Independent Speaker Verification against GMM-based Systems, EUROSPEECH 2001-SCANDINAVIA, 3, 1753