mlsanityData to 7 Oct 2026

Audio and omni-modal token compression

32 papers0 this weekHigh confidence in this grouping: the papers form one clear topicHigh confidence in this grouping: the papers form one clear topic · About

Keywords: audio, token, speech, compression, tokens

Key facts. Audio and omni-modal token compression: 32 papers on arXiv since 2023 (data through 7 Oct 2026). 11 of its 32 mapped papers (34%) cite “OmniSIFT: Modality-Asymmetric Token Compression for Efficient Omni-modal Large Language Models” (2026, arXiv 2602.04804). 9 of its 32 mapped papers (28%) cite “OmniZip: Audio-Guided Dynamic Token Compression for Fast Omnimodal Large Language Models” (2025, arXiv 2511.14582).

What’s new

Nothing new since 24 Sep. Latest: 31 Aug 2026.

Foundations

Most cited by papers here

  1. OmniSIFT: Modality-Asymmetric Token Compression for Efficient Omni-modal Large Language Models2026cited by 11 here
  2. OmniZip: Audio-Guided Dynamic Token Compression for Fast Omnimodal Large Language Models2025cited by 9 here
  3. SpeechPrune: Context-aware Token Pruning for Speech Information Retrieval2024cited by 7 here
  4. Accelerating Transducers through Adjacent Token Merging2023cited by 6 here
  5. EchoingPixels: Aliasing-Resistant Joint Token Reduction for Audio-Visual LLMs2025cited by 3 here

Lineage

The papers this direction builds on, oldest first.

Each paper on the line cites the one before.

  1. 2021Learned Token Pruning for Transformerscited by 2 here
  2. 2022Token Merging: Your ViT But Fastercited by 13 here
  3. 2024LLaVA-PruMerge: Adaptive Token Reduction for Efficient Large Multimodal Modelscited by 9 here
  4. 2024DyCoke: Dynamic Compression of Tokens for Fast Video Large Language Modelscited by 9 here
  5. 2025HoliTom: Holistic Token Merging for Fast Video Large Language Modelscited by 7 here
  6. 2025Variation-aware Vision Token Dropping for Faster Large Vision-Language Modelscited by 3 here
  7. 2025StreamingTOM: Streaming Token Compression for Efficient Video Understandingcited by 3 here
  8. 2025OmniZip: Audio-Guided Dynamic Token Compression for Fast Omnimodal Large Language Modelscited by 9 here
  9. 2026OmniSIFT: Modality-Asymmetric Token Compression for Efficient Omni-modal Large Language Modelscited by 11 here
15 more it builds on, not in this chain
  1. 2023Accelerating Transducers through Adjacent Token Mergingcited by 6 here
  2. 2023Efficient Memory Management for Large Language Model Serving with PagedAttentioncited by 4 here
  3. 2024An Image is Worth 1/2 Tokens After Layer 2: Plug-and-Play Inference Acceleration for Large Vision-Language Modelscited by 15 here
  4. 2024SnapKV: LLM Knows What You are Looking for Before Generationcited by 4 here
  5. 2024Fit and Prune: Fast and Training-free Visual Token Pruning for Multi-modal Large Language Modelscited by 8 here
  6. 2024Beyond Text-Visual Attention: Exploiting Visual Cues for Effective Token Pruning in VLMscited by 3 here
  7. 2024SpeechPrune: Context-aware Token Pruning for Speech Information Retrievalcited by 7 here
  8. 2024PruneVid: Visual Token Pruning for Efficient Video Large Language Modelscited by 6 here
  9. 2025Stop Looking for Important Tokens in Multimodal Language Models: Duplication Matters Morecited by 3 here
  10. 2025Token Pruning in Multimodal Large Language Models: Are We Solving the Right Problem?cited by 3 here
  11. 2025STORM: Token-Efficient Long Video Understanding for Multimodal LLMscited by 4 here
  12. 2025Token Pruning in Audio Transformers: Optimizing Performance and Decoding Patch Importancecited by 6 here
  13. 2025Video Compression Commander: Plug-and-Play Inference Acceleration for Video Large Language Modelscited by 5 here
  14. 2025Mixing Importance with Diversity: Joint Optimization for KV Cache Compression in Large Vision-Language Modelscited by 3 here
  15. 2025EchoingPixels: Aliasing-Resistant Joint Token Reduction for Audio-Visual LLMscited by 3 here

The 24 most cited of 92.

Browse

27 of 32, ranked 6–32

  1. 6DASH: Dynamic Audio-Driven Semantic Chunking for Efficient Omnimodal Token Compression20262
  2. 7Keep What Audio Cannot Say: Context-Preserving Token Pruning for Omni-LLMs20262
  3. 8OmniDrop: Layer-wise Token Pruning for Omni-modal LLMs via Query-Guidance20262
  4. 9OmniSelect: Dynamic Modality-Aware Token Compression for Efficient Omni-modal Large Language Models20262
  5. 10AudioMarathon: A Comprehensive Benchmark for Long-Context Audio Understanding and Efficiency in Audio LLMs20252
  6. 11Towards Audio Token Compression in Large Audio Language Models20252
  7. 12VoxServe: Streaming-Centric Serving System for Speech Language Models20261
  8. 13FastAdaSP: Multitask-Adapted Efficient Inference for Large Speech Language Model20241
  9. 14HeadRouter: Dynamic Head-Weight Routing for Task-Adaptive Audio Token Pruning in Large Audio Language Models20260
  10. 15Locality Matters for Training-Free Audio Token Compression in Audio-Language Models20260
All 27
  1. 16MURMUR: An Efficient Inference System for Long-Form ASR20260
  2. 17TLDR: Compressing Audio Tokens for Efficient Autoregressive Text-to-Speech20260
  3. 18AVOC: Enhancing Hour-Level Audio-Video Understanding in Omni-Modal LLMs via Retrieval-Inspired Token Compression20260
  4. 19OmniFocus: Query-Guided Modality-Balanced Token Compression for Omni-Modal Large Language Models20260
  5. 20Out of Sight, Still in Mind: Token Compression for Omni-LLMs20260
  6. 21OmniScope: Modality-Decoupled Token Compression for Omnimodal Large Language Models20260
  7. 22Omni-Prune: Query-Aware Unified Token Pruning for Efficient Omnimodal Large Language Models20260
  8. 23OmniDelta: Skill-Driven Budget Allocation for Token Compression in OmniLLMs20260
  9. 24Allocation Before Ranking: Decoupled Token Compression for OmniLLMs20260
  10. 25OmniPack: Unified Token Compression for Efficient Omni-modal Large Language Models20260
  11. 26Aero Realtime: Fully Aligned Input-Output Streams for Low-Latency Streaming Multimodal Generation20260
  12. 27VoxZip: Semantic-Anchored Temporal KV Cache Compression for Long-Context Audio Inference20260
  13. 28Deferred Audio Pruning with Local Audio-Visual Dynamics for Omni-LLMs20260
  14. 29WnW: Waxing-and-Waning KV Cache for Long-Form Speech LLMs20260
  15. 30Stride-k Subsampling: Train-Free Audio Token Reduction for Whisper20260
  16. 31Spiralformer: Low Latency Encoder for Streaming Speech Recognition with Circular Layer Skipping and Early Exiting20250
  17. 32FastAST: Accelerating Audio Spectrogram Transformer via Token Merging and Cross-Model Knowledge Distillation20240