DocumentCode
3642758
Title
Text independent speaker recognition using LBG vector quantization
Author
Danko Komlen;Tomislav Lombarović;Mario Ogrizek-Tomaš;Denis Petek;Andrej Petković
Author_Institution
University of Zagreb, Faculty of Electrical Engineering and Computing, Unska 3, 10000 Zagreb, Republic of Croatia
fYear
2011
fDate
5/1/2011 12:00:00 AM
Firstpage
1652
Lastpage
1657
Abstract
There is a great need for a system that will, in the absence of other biometric data, be able to identify the person by voice. This paper describes a system based on LBG vector quantization and the k-NN classifier, while the features that were used are MFCC coefficients and energy of the sound signal. Based on the described approach the developed system was evaluated on two sets of speakers. The results obtained are encouraging, with an accuracy of more than 95%. The system was also evaluated for the case of interference in the voice signal transmission, and accuracy in this case ranges from 70% up to 85%.
Keywords
"Accuracy","Speaker recognition","Databases","Mel frequency cepstral coefficient","Support vector machine classification","Training","Speech"
Publisher
ieee
Conference_Titel
MIPRO, 2011 Proceedings of the 34th International Convention
Print_ISBN
978-1-4577-0996-8
Type
conf
Filename
5967326
Link To Document