# Uncovering shortcut learning in audio classifiers by discovering recurring concepts in temporal explanations

arXiv: https://arxiv.org/abs/2609.34030 · HTML: https://mlsanity.com/paper/2609.34030/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 25 of its 28 references that are in the map.
- First posted on arXiv 27 Sep 2026.

## Builds on (most cited first, 25 of 25)

- [Qwen2 Technical Report](https://mlsanity.com/paper/2407.10671/) (2024)
- [Shortcut Learning in Deep Neural Networks](https://mlsanity.com/paper/2004.07780/) (2020)
- [Interpretability Beyond Feature Attribution: Quantitative Testing with Concept Activation Vectors (TCAV)](https://mlsanity.com/paper/1711.11279/) (2017)
- [Distributionally Robust Neural Networks for Group Shifts: On the Importance of Regularization for Worst-Case Generalization](https://mlsanity.com/paper/1911.08731/) (2019)
- [Qwen2.5-Omni Technical Report](https://mlsanity.com/paper/2503.20215/) (2025)
- [Qwen3-Omni Technical Report](https://mlsanity.com/paper/2509.17765/) (2025)
- [Unmasking Clever Hans Predictors and Assessing What Machines Really Learn](https://mlsanity.com/paper/1902.10178/) (2019)
- [Kimi-Audio Technical Report](https://mlsanity.com/paper/2504.18425/) (2025)
- [Audio Flamingo 3: Advancing Audio Intelligence with Fully Open Large Audio Language Models](https://mlsanity.com/paper/2507.08128/) (2025)
- [Domino: Discovering Systematic Errors with Cross-Modal Embeddings](https://mlsanity.com/paper/2203.14960/) (2022)
- [The Benefit Of Temporally-Strong Labels In Audio Event Classification](https://mlsanity.com/paper/2105.07031/) (2021)
- [MAE-AST: Masked Autoencoding Audio Spectrogram Transformer](https://mlsanity.com/paper/2203.16691/) (2022)
- [A Multimodal Automated Interpretability Agent](https://mlsanity.com/paper/2404.14394/) (2024)
- [Navigating Shortcuts, Spurious Correlations, and Confounders: From Origins via Detection to Mitigation](https://mlsanity.com/paper/2412.05152/) (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)
- [Music Flamingo: Scaling Music Understanding in Audio Language Models](https://mlsanity.com/paper/2511.10289/) (2025)
- [Audio Flamingo Next: Next-Generation Open Audio-Language Models for Speech, Sound, and Music](https://mlsanity.com/paper/2604.10905/) (2026)
- [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)
- [Concept-Based Techniques for "Musicologist-friendly" Explanations in a Deep Music Classifier](https://mlsanity.com/paper/2208.12485/) (2022)
- [Shortcut Learning in Binary Classifier Black Boxes: Applications to Voice Anti-Spoofing and Biometrics](https://mlsanity.com/paper/2601.17782/) (2026)
- [LMAC-TD: Producing Time Domain Explanations for Audio Classifiers](https://mlsanity.com/paper/2409.08655/) (2024)
- [Benchmarking Time-localized Explanations for Audio Classification Models](https://mlsanity.com/paper/2506.04391/) (2025)
- [An Audio Language Model-Based Voice Concept Bottleneck Framework for Interpretable Health Assessment](https://mlsanity.com/paper/2607.16967/) (2026)

## Abstract (arXiv)

Correlations between events in machine learning datasets may result in shortcut learning, where models learn to predict the target event based on the presence of a correlated event. When these correlations are spurious -- arising from data collection artifacts -- models are likely to perform poorly in practice. We propose a pipeline to uncover shortcut learning in audio classifiers by discovering recurring concepts in their temporal explanations. Specifically, we isolate audio segments that explain classifier decisions, caption them with an ensemble of Large Audio-Language Models, and use a Large Language Model to extract recurring concepts. The resulting concepts can be audited by humans to uncover potential shortcut learning. We evaluate our framework using datasets curated from AudioSet Strong, controlling for the presence or absence of spurious correlations. Results show that this approach reliably uncovers learned shortcuts, such as the model relying on the presence of "laughter" to predict "applause".