What Does an Audio Deepfake Detector Focus on? A Study in the Time Domain

TL;DR

This study introduces a relevancy-based explainable AI method for audio deepfake detection, demonstrating its effectiveness over standard methods and revealing insights into model focus across large datasets.

Contribution

The paper proposes a novel relevancy-based XAI approach for transformer-based audio deepfake detectors and evaluates its performance on large datasets, highlighting differences from existing methods.

Findings

Relevancy-based XAI outperforms Grad-CAM and SHAP in faithfulness metrics.

Model explanations vary significantly between limited and large datasets.

Analysis suggests speech/non-speech and phonetic content influence model focus.

Abstract

Adding explanations to audio deepfake detection (ADD) models will boost their real-world application by providing insight on the decision making process. In this paper, we propose a relevancy-based explainable AI (XAI) method to analyze the predictions of transformer-based ADD models. We compare against standard Grad-CAM and SHAP-based methods, using quantitative faithfulness metrics as well as a partial spoof test, to comprehensively analyze the relative importance of different temporal regions in an audio. We consider large datasets, unlike previous works where only limited utterances are studied, and find that the XAI methods differ in their explanations. The proposed relevancy-based XAI method performs the best overall on a variety of metrics. Further investigation on the relative importance of speech/non-speech, phonetic content, and voice onsets/offsets suggest that the XAI results obtained from analyzing limited utterances don't necessarily hold when evaluated on large datasets.