eduzhai > Applied Sciences > Engineering >

Using Multi-Resolution Feature Maps with Convolutional Neural Networks for Anti-Spoofing in ASV

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 5 pages

Abstract: This paper presents a simple but effective method that uses multi-resolutionfeature maps with convolutional neural networks (CNNs) for anti-spoofing inautomatic speaker verification (ASV). The central idea is to alleviate theproblem that the feature maps commonly used in anti-spoofing networks areinsufficient for building discriminative representations of audio segments, asthey are often extracted by a single-length sliding window. Resultingtrade-offs between time and frequency resolutions restrict the information insingle spectrograms. The proposed method improves both frequency resolution andtime resolution by stacking multiple spectrograms that are extracted usingdifferent window lengths. These are fed into a convolutional neural network inthe form of multiple channels, making it possible to extract more informationfrom input signals while only marginally increasing computational costs. Theefficiency of the proposed method has been conformed on the ASVspoof 2019database. We show that the use of the proposed multiresolution inputsconsistently outperforms that of score fusion across different CNNarchitectures. Moreover, computational cost remains small.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...