A Framework for Evaluating Faithfulness in Explainable AI for Machine Anomalous Sound Detection Using Frequency-Band Perturbation

TL;DR

This paper introduces a quantitative framework to evaluate the faithfulness of explainable AI methods in machine anomalous sound detection by systematically removing frequency bands and assessing model response, revealing reliability differences among methods.

Contribution

The authors propose a novel, reproducible framework for objectively benchmarking audio explanation methods based on frequency-band removal and model sensitivity analysis.

Findings

Occlusion method shows the strongest alignment with true model sensitivity.

Gradient-based methods often fail to accurately capture spectral dependencies.

The framework enables trustworthy interpretation of spectrogram-based ASD systems.

Abstract

Explainable AI (XAI) is commonly applied to anomalous sound detection (ASD) models to identify which time-frequency regions of an audio signal contribute to an anomaly decision. However, most audio explanations rely on qualitative inspection of saliency maps, leaving open the question of whether these attributions accurately reflect the spectral cues the model uses. In this work, we introduce a new quantitative framework for evaluating XAI faithfulness in machine-sound analysis by directly linking attribution relevance to model behaviour through systematic frequency-band removal. This approach provides an objective measure of whether an XAI method for machine ASD correctly identifies frequency regions that influence an ASD model's predictions. By using four widely adopted methods, namely Integrated Gradients, Occlusion, Grad-CAM and SmoothGrad, we show that XAI techniques differ in reliability, with Occlusion demonstrating the strongest alignment with true model sensitivity and gradient-+based methods often failing to accurately capture spectral dependencies. The proposed framework offers a reproducible way to benchmark audio explanations and enables more trustworthy interpretation of spectrogram-based ASD systems.