Konuşmacı tanıma sistemleri için güvenlik hayati önem taşımaktadır. Geçtiğimiz yıllarda, sahte konuşma saldırılarının bu sistemleri kandırabildiği ortaya konmuştur. Bu durumu önlemek amacı ile sahte konuşma tespit sistemleri geliştirilmiştir. Bu tür sistemler bazı durumlarda oldukça yüksek performans sergilese de, gürültü altında performansları kötüleşmektedir. Geleneksel konuşma iyileştirme yöntemleri performansı artırmak bir yana, daha da kötüleştirmektedir. Bu çalışmada, konvolüsyonel sinir ağı yapısı kullanılarak elde edilen maskenin gürültü etkisini azaltmaktaki performansı incelenmiştir. Maske, spektrogramın gürültülü bölgelerini bastırmakta ve bu spektrogramdan elde edilen i-vectorleri gürbüz hale getirmekte kullanılmıştır. ASVspoof 2015 veri tabanı ve üç farklı gürültü tipi ile gerçekleştirilen testlerde önerilen sistemin geleneksel sistemlerden daha üstün olduğu gösterilmiştir. Ancak eğitim aşamasında karşılaşılmayan gürültü tiplerinde performans kaybı olmaktadır.
derin öğrenme evrişimli sinir ağı sahte konuşma tanıma konuşmacı tanıma gürbüz öznitelikler
121E057
Ensuring security in speaker recognition systems is crucial. In the past years, it has been demonstrated that spoofing attacks can fool these systems. In order to deal with this issue, spoof speech detection systems have been developed. While these systems have served with a good performance, their effectiveness tends to degrade under noise. Traditional speech enhancement methods are not efficient for improving performance, they even make it worse. In this research paper, performance of the noise mask obtained via a convolutional neural network structure for reducing the noise effects was investigated. The mask is used to suppress noisy regions of spectrograms in order to extract robust i-vectors. The proposed system is tested on the ASVspoof 2015 database with three different noise types and accomplished superior performance compared to the traditional systems. However, there is a loss of performance in noise types that are not encountered during training phase.
deep learning convolutional neural network spoof detection speaker recognition robust features
TÜBİTAK
121E057
This work was supported by TÜBİTAK (Project No: 121E057).
Primary Language | English |
---|---|
Subjects | Software Engineering (Other) |
Journal Section | Research Articles |
Authors | |
Project Number | 121E057 |
Early Pub Date | March 28, 2024 |
Publication Date | April 22, 2024 |
Submission Date | June 7, 2023 |
Acceptance Date | March 15, 2024 |
Published in Issue | Year 2024 Volume: 29 Issue: 1 |
Announcements:
30.03.2021-Beginning with our April 2021 (26/1) issue, in accordance with the new criteria of TR-Dizin, the Declaration of Conflict of Interest and the Declaration of Author Contribution forms fulfilled and signed by all authors are required as well as the Copyright form during the initial submission of the manuscript. Furthermore two new sections, i.e. ‘Conflict of Interest’ and ‘Author Contribution’, should be added to the manuscript. Links of those forms that should be submitted with the initial manuscript can be found in our 'Author Guidelines' and 'Submission Procedure' pages. The manuscript template is also updated. For articles reviewed and accepted for publication in our 2021 and ongoing issues and for articles currently under review process, those forms should also be fulfilled, signed and uploaded to the system by authors.