Complex-valued neural networks for voice anti-spoofing

TL;DR

This paper introduces complex-valued neural networks that process complex frequency representations of audio, retaining phase information and improving voice anti-spoofing detection while enabling explainability.

Contribution

It proposes a novel complex-valued neural network approach that combines magnitude and phase information for improved anti-spoofing detection and interpretability.

Findings

Outperforms previous methods on the 'In-the-Wild' dataset

Retains phase information for better audio naturalness

Enables explainable AI techniques for model interpretation

Abstract

Current anti-spoofing and audio deepfake detection systems use either magnitude spectrogram-based features (such as CQT or Melspectrograms) or raw audio processed through convolution or sinc-layers. Both methods have drawbacks: magnitude spectrograms discard phase information, which affects audio naturalness, and raw-feature-based models cannot use traditional explainable AI methods. This paper proposes a new approach that combines the benefits of both methods by using complex-valued neural networks to process the complex-valued, CQT frequency-domain representation of the input audio. This method retains phase information and allows for explainable AI methods. Results show that this approach outperforms previous methods on the "In-the-Wild" anti-spoofing dataset and enables interpretation of the results through explainable AI. Ablation studies confirm that the model has learned to use phase information to detect voice spoofing.