# Mask-Induced Displacement in Audio XAI via Logit Trajectory Decomposition

arXiv: https://arxiv.org/abs/2609.33486 · HTML: https://mlsanity.com/paper/2609.33486/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 22 of its 31 references that are in the map.
- First posted on arXiv 27 Sep 2026.

## Builds on (most cited first, 22 of 22)

- [A Unified Approach to Interpreting Model Predictions](https://mlsanity.com/paper/1705.07874/) (2017)
- ["Why Should I Trust You?": Explaining the Predictions of Any Classifier](https://mlsanity.com/paper/1602.04938/) (2016)
- [SpecAugment: A Simple Data Augmentation Method for Automatic Speech Recognition](https://mlsanity.com/paper/1904.08779/) (2019)
- [RISE: Randomized Input Sampling for Explanation of Black-box Models](https://mlsanity.com/paper/1806.07421/) (2018)
- [AST: Audio Spectrogram Transformer](https://mlsanity.com/paper/2104.01778/) (2021)
- [Interpretable Explanations of Black Boxes by Meaningful Perturbation](https://mlsanity.com/paper/1704.03296/) (2017)
- [PANNs: Large-Scale Pretrained Audio Neural Networks for Audio Pattern Recognition](https://mlsanity.com/paper/1912.10211/) (2019)
- [Explaining by Removing: A Unified Framework for Model Explanation](https://mlsanity.com/paper/2011.14878/) (2020)
- [The Disagreement Problem in Explainable Machine Learning: A Practitioner's Perspective](https://mlsanity.com/paper/2202.01602/) (2022)
- [A Consistent and Efficient Evaluation Strategy for Attribution Methods](https://mlsanity.com/paper/2202.00449/) (2022)
- [Missingness Bias in Model Debugging](https://mlsanity.com/paper/2204.08945/) (2022)
- [Order in the Court: Explainable AI Methods Prone to Disagreement](https://mlsanity.com/paper/2105.03287/) (2021)
- [Decoupling Pixel Flipping and Occlusion Strategy for Consistent XAI Benchmarks](https://mlsanity.com/paper/2401.06654/) (2024)
- [Listen to Interpret: Post-hoc Interpretability for Audio Networks with NMF](https://mlsanity.com/paper/2202.11479/) (2022)
- [audioLIME: Listenable Explanations Using Source Separation](https://mlsanity.com/paper/2008.00582/) (2020)
- [Reliable Local Explanations for Machine Listening](https://mlsanity.com/paper/2005.07788/) (2020)
- [Listenable Maps for Audio Classifiers](https://mlsanity.com/paper/2403.13086/) (2024)
- [Explanations for Automatic Speech Recognition](https://mlsanity.com/paper/2302.14062/) (2023)
- [From Flexibility to Manipulation: The Slippery Slope of XAI Evaluation](https://mlsanity.com/paper/2412.05592/) (2024)
- [Exploring Local Interpretable Model-Agnostic Explanations for Speech Emotion Recognition with Distribution-Shift](https://mlsanity.com/paper/2504.05368/) (2025)
- [I Guess That's Why They Call it the Blues: Causal Analysis for Audio Classifiers](https://mlsanity.com/paper/2601.16675/) (2026)
- [If It's Good Enough for You, It's Good Enough for Me: Transferability of Audio Sufficiencies across Models](https://mlsanity.com/paper/2604.02937/) (2026)

## Abstract (arXiv)

Perturbation-based XAI methods for audio classifiers often estimate feature importance by masking spectrogram regions and crediting output changes to the retained signal. Yet they typically assume the fill (the mask replacement) is negligible. We propose logit-space trajectory decomposition to examine this assumption. An on-axis component captures output along a line connecting the fully filled (occluded) spectrogram to the fully retained original; an off-axis component captures perpendicular displacement. We evaluate across three fills, three audio classifiers, and 1,100 AudioSet clips. We demonstrate that no fill is acoustically neutral under full occlusion: Zero fill activates silence and Gaussian Noise fill activates broadband noise. Under partial masking, 41-77% of output displacement is off-axis, with the direction of the residual stable across mask retention fraction and specific to each model-fill combination. Attribution methods are unevenly exposed to off-axis displacement through their sampling and weighting strategies, revealing apparatus-dependence. Where the off-axis residual is stable and low-dimensional, its structure affords mitigation.