D-vector based speaker verification system using Raw Waveform CNN

DOI

10.2991/anit-17.2018.21How to use a DOI?

Keywords

d-vector, speaker verification, raw-audio-CNN

Abstract

In this paper, we propose a d-vector based speaker verification system in which raw-audio-CNN is used as a d-vector extractor instead of a conventional multi-layer perceptron. Because raw-audio-CNN takes raw wave signals as input, traditional acoustic feature extracting methods such as mel-frequency cepstral coefficient and mel-filterbank features are no longer needed. The results of experiments conducted show that raw-audio-CNN can successfully perform functions carried out by traditional acoustic feature extracting methods and outperforms traditional d-vector systems that utilize standard multi-layer perceptron with acoustic features.

Copyright

© 2018, the Authors. Published by Atlantis Press.

Open Access

This is an open access article distributed under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/).

Download article (PDF)

TY  - CONF
AU  - Jeeweon Jung
AU  - Heesoo Heo
AU  - Ilho Yang
AU  - Sunghyun Yoon
AU  - Hyejin Shim
AU  - Hajin Yu
PY  - 2017/12
DA  - 2017/12
TI  - D-vector based speaker verification system using Raw Waveform CNN
BT  - Proceedings of the 2017 International Seminar on Artificial Intelligence, Networking and Information Technology (ANIT 2017)
PB  - Atlantis Press
SP  - 126
EP  - 131
SN  - 1951-6851
UR  - https://doi.org/10.2991/anit-17.2018.21
DO  - 10.2991/anit-17.2018.21
ID  - Jung2017/12
ER  -