Deep Learning Approach: Detection of Replay Attack in ASV Systems
| Autor: | S. Saranya, Suvidha Rupesh Kumar, B. Bharathi |
|---|---|
| Rok vydání: | 2020 |
| Předmět: |
0209 industrial biotechnology
Spoofing attack Artificial neural network Computer science business.industry Speech recognition Deep learning 02 engineering and technology Convolutional neural network 020901 industrial engineering & automation 0202 electrical engineering electronic engineering information engineering 020201 artificial intelligence & image processing Artificial intelligence Mel-frequency cepstrum business Classifier (UML) Replay attack |
| Zdroj: | Advances in Intelligent Systems and Computing ISBN: 9789811524745 |
| DOI: | 10.1007/978-981-15-2475-2_27 |
| Popis: | Automatic speaker verification (ASV) system is a bio-metric authentication system, which accepts or rejects speech utterance depending on the voiceprint of speakers. Vulnerability of such system to spoofing attacks is the biggest challenge at hand. Proposed system is an initiative toward improvising the countermeasure for replay attacks on ASV systems. It is developed and tested on ASVspoof2017 corpus. Constant Q cepstral coefficients (CQCCs) feature is extracted to build the models using simple neural network and convolutional neural network. CQCC feature is chosen, as it a feature proved to be robust for ASV spoof detection when tested on GMM classifier. A significant improvement of 4.9% EER is achieved using convolution neural network (CNN). |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|