Paper 2602.05027
AudioSAE: Towards Understanding of Audio-Processing Models with Sparse AutoEncoders
- Published
- Feb 2026
- Research lab
- Independent
- Citations
- 5
- GitHub
- 15 stars
01 In brief
Summary
This paper introduces AudioSAE, the first large-scale application of Sparse Autoencoders (SAEs) to audio models, training them on all encoder layers of Whisper and HuBERT.
The authors evaluate feature stability, interpretability, and practical utility.
Over 50% of features remain consistent across random seeds, and reconstruction quality is preserved.
SAE features capture general acoustic and semantic information as well as specific events like environmental noises and paralinguistic sounds (e.g., laughter, whispering), and they effectively disentangle concepts, requiring removal of only 19–27% of features to erase a concept.
Feature steering reduces Whisper's false speech detections by 70% with negligible WER increase.
Additionally, some SAE features correlate with human EEG activity during speech perception, indicating alignment with neural processing.
The code and checkpoints are publicly available.
02 From the paper
Abstract
Sparse Autoencoders (SAEs) are powerful tools for interpreting neural representations, yet their use in audio remains underexplored. We train SAEs across all encoder layers of Whisper and HuBERT, provide an extensive evaluation of their stability, interpretability, and show their practical utility. Over 50% of the features remain consistent across random seeds, and reconstruction quality is preserved. SAE features capture general acoustic and semantic information as well as specific events, including environmental noises and paralinguistic sounds (e.g. laughter, whispering) and disentangle them effectively, requiring removal of only 19-27% of features to erase a concept. Feature steering reduces Whisper's false speech detections by 70% with negligible WER increase, demonstrating real-world applicability. Finally, we find SAE features correlated with human EEG activity during speech perception, indicating alignment with human neural processing. The code and checkpoints are available at https://github.com/audiosae/audiosae_demo.