# Token Pruning in Audio Transformers: Optimizing Performance and Decoding Patch Importance

arXiv: https://arxiv.org/abs/2504.01690 · HTML: https://mlsanity.com/paper/2504.01690/

## Key facts

- Cited by 12 papers in the map (data through 14 Oct 2026).
- Placed in Token pruning and merging for ViTs, one of 12 directions in Efficient LLM inference and serving.
- Builds on 34 of its 41 references that are in the map.
- First posted on arXiv 2 Apr 2025.

## Placed in

- Efficient LLM inference and serving / [Token pruning and merging for ViTs](https://mlsanity.com/inference/token-pruning-and-merging-for-vits/)

## Builds on (most cited first, 30 of 34)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://mlsanity.com/paper/2010.11929/) (2020)
- [PyTorch: An Imperative Style, High-Performance Deep Learning Library](https://mlsanity.com/paper/1912.01703/) (2019)
- [Swin Transformer: Hierarchical Vision Transformer using Shifted Windows](https://mlsanity.com/paper/2103.14030/) (2021)
- [Masked Autoencoders Are Scalable Vision Learners](https://mlsanity.com/paper/2111.06377/) (2021)
- [SpecAugment: A Simple Data Augmentation Method for Automatic Speech Recognition](https://mlsanity.com/paper/1904.08779/) (2019)
- [Mixed Precision Training](https://mlsanity.com/paper/1710.03740/) (2017)
- [Token Merging: Your ViT But Faster](https://mlsanity.com/paper/2210.09461/) (2022)
- [Multiscale Vision Transformers](https://mlsanity.com/paper/2104.11227/) (2021)
- [Speech Commands: A Dataset for Limited-Vocabulary Speech Recognition](https://mlsanity.com/paper/1804.03209/) (2018)
- [AST: Audio Spectrogram Transformer](https://mlsanity.com/paper/2104.01778/) (2021)
- [MaxViT: Multi-Axis Vision Transformer](https://mlsanity.com/paper/2204.01697/) (2022)
- [Not All Patches are What You Need: Expediting Vision Transformers via Token Reorganizations](https://mlsanity.com/paper/2202.07800/) (2022)
- [AdaViT: Adaptive Tokens for Efficient Vision Transformer](https://mlsanity.com/paper/2112.07658/) (2021)
- [Post-Training Quantization for Vision Transformer](https://mlsanity.com/paper/2106.14156/) (2021)
- [Masked Autoencoders that Listen](https://mlsanity.com/paper/2207.06405/) (2022)
- [Adaptive Token Sampling For Efficient Vision Transformers](https://mlsanity.com/paper/2111.15667/) (2021)
- [HTS-AT: A Hierarchical Token-Semantic Audio Transformer for Sound Classification and Detection](https://mlsanity.com/paper/2202.00874/) (2022)
- [SSAST: Self-Supervised Audio Spectrogram Transformer](https://mlsanity.com/paper/2110.09784/) (2021)
- [SPViT: Enabling Faster Vision Transformers via Soft Token Pruning](https://mlsanity.com/paper/2112.13890/) (2021)
- [Token Fusion: Bridging the Gap between Token Pruning and Token Merging](https://mlsanity.com/paper/2312.01026/) (2023)
- [Which Tokens to Use? Investigating Token Reduction in Vision Transformers](https://mlsanity.com/paper/2308.04657/) (2023)
- [EAT: Self-Supervised Pre-Training with Efficient Audio Transformer](https://mlsanity.com/paper/2401.03497/) (2024)
- [Multi-Exit Vision Transformer for Dynamic Inference](https://mlsanity.com/paper/2106.15183/) (2021)
- [Self-supervised Audio Teacher-Student Transformer for Both Clip-level and Frame-level Tasks](https://mlsanity.com/paper/2306.04186/) (2023)
- [Dynamic Spatial Sparsification for Efficient Vision Transformers and Convolutional Neural Networks](https://mlsanity.com/paper/2207.01580/) (2022)
- [Masked Modeling Duo: Towards a Universal Audio Pre-training Framework](https://mlsanity.com/paper/2404.06095/) (2024)
- [Learned Thresholds Token Merging and Pruning for Vision Transformers](https://mlsanity.com/paper/2307.10780/) (2023)
- [SpeechPrune: Context-aware Token Pruning for Speech Information Retrieval](https://mlsanity.com/paper/2412.12009/) (2024)
- [HuBERT-EE: Early Exiting HuBERT for Efficient Speech Recognition](https://mlsanity.com/paper/2204.06328/) (2022)

## Built on by (most cited first, 12 of 12)

- [A Survey of Token Compression for Efficient Multimodal Large Language Models](https://mlsanity.com/paper/2507.20198/) (2025)
- [OmniZip: Audio-Guided Dynamic Token Compression for Fast Omnimodal Large Language Models](https://mlsanity.com/paper/2511.14582/) (2025)
- [FastAV: Efficient Token Pruning for Audio-Visual Large Language Model Inference](https://mlsanity.com/paper/2601.13143/) (2026)
- [Do We Need Distinct Representations for Every Speech Token? Unveiling and Exploiting Redundancy in Large Speech Language Models](https://mlsanity.com/paper/2604.06871/) (2026)
- [OmniRefine: Alignment-Aware Cooperative Compression for Efficient Omnimodal Large Language Models](https://mlsanity.com/paper/2605.12056/) (2026)
- [OmniPack: Unified Token Compression for Efficient Omni-modal Large Language Models](https://mlsanity.com/paper/2608.03812/) (2026)
- [OmniFocus: Query-Guided Modality-Balanced Token Compression for Omni-Modal Large Language Models](https://mlsanity.com/paper/2607.03050/) (2026)
- [OmniScope: Modality-Decoupled Token Compression for Omnimodal Large Language Models](https://mlsanity.com/paper/2607.23193/) (2026)
- [OmniDelta: Skill-Driven Budget Allocation for Token Compression in OmniLLMs](https://mlsanity.com/paper/2607.25669/) (2026)
- [The silence of the weights: a structural pruning strategy for attention-based audio signal architectures with second order metrics](https://mlsanity.com/paper/2509.26207/) (2025)
- [Stride-k Subsampling: Train-Free Audio Token Reduction for Whisper](https://mlsanity.com/paper/2608.30927/) (2026)
- [Audio Token Attention Is Predictable Before the Language Model Runs](https://mlsanity.com/paper/2609.38878/) (2026)

## Abstract (arXiv)

Vision Transformers (ViTs) have achieved state-of-the-art performance across various computer vision tasks, but their high computational cost remains a challenge. Token pruning has been proposed to reduce this cost by selectively removing less important tokens. While effective in vision tasks by discarding non-object regions, applying this technique to audio tasks presents unique challenges, as distinguishing relevant from irrelevant regions in time-frequency representations is less straightforward. In this study, for the first time, we applied token pruning to ViT-based audio classification models using Mel-spectrograms and analyzed the trade-offs between model performance and computational cost: TopK token pruning can reduce MAC operations of AudioMAE and AST by 30-40%, with less than a 1% drop in accuracy. Our analysis reveals that while high-intensity or high-variation tokens contribute significantly to model accuracy, low-intensity or low variation tokens also remain important when token pruning is applied; pruning solely based on the intensity or variation of signals in a patch leads to a noticeable drop in accuracy. We support our claim by measuring high correlation between attention scores and these statistical features and by showing retained tokens consistently receive distinct attention compared to pruned ones. We also show that AudioMAE retains more low-intensity tokens than AST. This can be explained by AudioMAE's self-supervised reconstruction objective, which encourages attention to all patches, whereas AST's supervised training focuses on label-relevant tokens.