Home >

news Help

Publication Information


Title
Japanese: 
English:I-vector Transformation Using Conditional Generative Adversarial Networks for Short Utterance Speaker Verification 
Author
Japanese: ZHANG Jiacen, 井上 中順, 篠田 浩一.  
English: Jiacen Zhang, Nakamasa Inoue, Koichi Shinoda.  
Language English 
Journal/Book name
Japanese: 
English:Proc. Interspeech 2018 
Volume, Number, Page         pp. 3613-3617
Published date Sep 4, 2018 
Publisher
Japanese: 
English:ISCA 
Conference name
Japanese: 
English:Interspeech 2018 
Conference site
Japanese:ハイデラバード 
English:Hyderabad 
File
Official URL https://www.isca-speech.org/archive/Interspeech_2018/abstracts/1680.html
 
DOI https://doi.org/10.21437/Interspeech.2018-1680
Abstract I-vector based text-independent speaker verification (SV) systems often have poor performance with short utterances, as the biased phonetic distribution in a short utterance makes the extracted i-vector unreliable. This paper proposes an i-vector compensation method using a generative adversarial network (GAN), where its generator network is trained to generate a compensated i-vector from a short-utterance i-vector and its discriminator network is trained to determine whether an i-vector is generated by the generator or the one extracted from a long utterance. Additionally, we assign two other learning tasks to the GAN to stabilize its training and to make the generated i-vector more speaker-specific. Speaker verification experiments on the NIST SRE 2008 “10sec-10sec” condition show that after applying our method, the equal error rate reduced by 11.3% from the conventional i-vector and PLDA system.

©2007 Tokyo Institute of Technology All rights reserved.