mlsanityData to 19 Oct 2026

AudioScorer: Learning Question-Aware Token Relevance for Large Audio Language Models

arXiv 2610.26258 ↗First posted 19 Oct 2026cs.SD cs.CL eess.AScited by 0

On the map: Speech and audio language models → Audio language models and spoken dialogue

Abstract (arXiv)

Large Audio Language Models (LALMs) reason over speech, environmental sounds, music, and other acoustic information. However, their audio encoders produce dense token sequences that make long-form audio costly to process. Existing audio pruning methods reduce this cost using heuristics such as attention, representation similarity, or temporal sampling, but do not explicitly learn which tokens are most useful for a given question. We propose AudioScorer, a lightweight model that scores audio tokens according to the user question. It is trained by distilling the ranking of audio tokens under the frozen LALM's answer-gradient saliency, together with downstream answer supervision. Compared with other pruning methods, AudioScorer achieves the best or tied-best result at 10% token retention on 7 of 10 semantic and acoustic tasks, including speech recognition, emotion recognition, and audio-scene classification.

Key facts. No citing papers in the map yet (data through 19 Oct 2026). Placed in Audio language models and spoken dialogue, one of 11 directions in Speech and audio language models. Builds on 12 of its 26 references that are in the map. First posted on arXiv 19 Oct 2026.

Builds on

12 of 26 references.

Closest ancestors · 2

Same direction, newest first

  1. 2025AudioMarathon: A Comprehensive Benchmark for Long-Context Audio Understanding and Efficiency in Audio LLMs8 Oct
  2. 2024Let's Go Real Talk: Spoken Dialogue Model for Face-to-Face Conversation12 Jun

Other references · 10

  1. 2023Lost in the Middle: How Language Models Use Long Contextscited by 3,519
  2. 2013Estimating or Propagating Gradients Through Stochastic Neurons for Conditional Computationcited by 2,149
  3. 2022Token Merging: Your ViT But Fastercited by 1,050
  4. 2023C-Pack: Packed Resources For General Chinese Embeddingscited by 604
  5. 2016YouTube-8M: A Large-Scale Video Classification Benchmarkcited by 334
  6. 2025Stop Looking for Important Tokens in Multimodal Language Models: Duplication Matters Morecited by 165
  7. 2023Mustango: Toward Controllable Text-to-Music Generationcited by 106
  8. 2025MMAU-Pro: A Challenging and Comprehensive Benchmark for Holistic Evaluation of Audio General Intelligencecited by 81
  9. 2022Audio Retrieval with WavText5K and CLAP Trainingcited by 43
  10. 2026Do We Need Distinct Representations for Every Speech Token? Unveiling and Exploiting Redundancy in Large Speech Language Modelscited by 5

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Enhancing Audio Reasoning via Semantic Summary Prediction
  2. 2026Omni-Prune: Query-Aware Unified Token Pruning for Efficient Omnimodal Large Language Models
  3. 2024Enhancing Temporal Understanding in Audio Question Answering for Large Audio Language Models
  4. 2026AudioDER: A Deduplication-Enhanced Reasoning Dataset for Post-Training Large Audio-Language Models
  5. 2025Segmentwise Pruning in Audio-Language Models
  6. 2026Task-Aware Answer Preservation under Audio Compression for Large Audio Language Models
  7. 2026HeadRouter: Dynamic Head-Weight Routing for Task-Adaptive Audio Token Pruning in Large Audio Language Models
  8. 2024SpeechPrune: Context-aware Token Pruning for Speech Information Retrieval