Toward Robust Audio Spoofing Detection: A Detailed Comparison of Traditional and Learned Features

Abstract

Automatic speaker verification, like every other biometric system, is vulnerable to spoofing attacks. Using only a few minutes of recorded voice of a genuine client of a speaker verification system, attackers can develop a variety of spoofing attacks that might trick such systems. Detecting these attacks using the audio cues present in the recordings is an important challenge. Most existing spoofing detection systems depend on knowing the used spoofing technique. With this research, we aim at overcoming this limitation, by examining robust audio features, both traditional and those learned through an autoencoder, that are generalizable to different types of replay spoofing. Furthermore, we provide a detailed account of all the steps necessary in setting up state-of-the-art audio feature detection, pre-, and postprocessing, such that the (non-audio expert) machine learning researcher can implement such systems. Finally, we evaluate the performance of our robust replay spoofing detection system with a wide variety and different combinations of both extracted and machine learned audio features on the 'out in the wild' ASVspoof 2017 dataset. This dataset contains a variety of new replay spoofing configurations. Since our focus is on examining which features will ensure robustness, we base our system on a traditional Gaussian Mixture Model-Universal Background Model (GMM-UBM). We then systematically investigate the relative contribution of each feature set. The fused models, based on both the known audio features and the machine learned features respectively, have a comparable performance with an Equal Error Rate (EER) of 12. The final best performing model, which obtains an EER of 10.8, is a hybrid system that contains both known and machine learned features, and is trained on an augmented dataset, thus revealing the importance of incorporating both types of features when developing a robust spoofing prediction model.

High-Performance Fake Voice Detection on Automatic Speaker Verification Systems for the Prevention of Cyber Fraud with Convolutional Neural Networks

“…Meanwhile, the false-negative rate (i.e., false alarm rate) refers to the ratio between misclassified genuine test samples and the total number of genuine test samples. Therefore, the lower the EER value, the better the developed countermeasure performs [58].…”

Deep Neural Network-Based Respiratory Pathology Classification Using Cough Sounds

“…The pre-processed audio signals were first segmented into frames of 100 ms, after which a Hamming window was applied, followed by the extraction of audio features. Mel-Frequency Cepstral Coefficients (MFCCs) were chosen for this investigation owing to their effectiveness when it comes to audio classification problems [34,35]. MFCCs are a set of features that focus on the perceptually relevant aspects of the audio spectrum, additionally the coefficients could contain information about the vocal tract characteristics [36,37].…”

Static–dynamic features and hybrid deep learning models based spoof detection system for ASV

“…have performed remarkably well for the spoof detection tasks, and for speech and speaker recognition tasks as well. These techniques can model the human vocal tract and human auditory system very well [15][16][17]. Human ear is proved to be deaf for the phase factor of sound.…”