[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-86326-en":3,"doc-seo-86326-105":29,"detail-sidebar-cat-0-en-105":91},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":13,"seo_description":14,"update_tm":27,"read_time":28},86326,7971461741311,"Ophelia","https://ap-avatar.wpscdn.com/avatar/74000253aff267980c6?x-image-process=image/resize,m_fixed,w_180,h_180&k=1779345379180704826",6,"Technology","Encoder-Side Neuron Identification and Amplification for Acoustic Perception in Large Audio-Language Models","Large audio-language models often understand speech content but underperform on fine-grained, non-semantic attributes such as a speaker’s emotion, especially because existing inference-time interventions typically act only after the audio encoder and at coarse granularity. IAAN (Identifying and Amplifying Acoustic Neurons) is a training-free, label-free method that scores individual audio-encoder neurons by contrasting activations on real waveforms versus noise references, then amplifies the highest-scoring neurons at inference. Experiments on ten attributes show large accuracy gains across multiple models.","Encoder-Side Neuron Identification and Amplification for Acoustic Perception in Large Audio-Language Models  \nYu-Han Huang 1,4 , Chih-Kai Yang2,* , Ke-Han Lu2,* , An-Yu Cheng 1,* , Hung-yi Lee3  \n1National Taiwan University, Taiwan  \n2 Graduate Institute of Communication Engineering, National Taiwan University, Taiwan  \n3NTU Artificial Intelligence Center of Research Excellence (NTU AI-CoRE), Taiwan  \n4ASUS Open Cloud Infrastructure Software Center, Taipei, Taiwan  \narXiv :2607 . 1 180 1v 1 [ cs . SD] 13 Jul 2026  \nAbstract—Large audio-language models (LALMs) often underperform on fine-grained, non-semantic attributes of speech, such as a speaker’semotion, despite strong performance on speech content. Improving this without the cost of retraining calls for an effective inference-time intervention, yet most existing methods intervene only after the audio encoder and operate at a relatively coarse granularity. The encoder itself, where acoustic information is first extracted from the waveform, remains largely unexplored, especially at the level of individual neurons. To this end, we introduce IAAN, Identifying and Amplifying Acoustic Neurons, a training-free and label-free method that scores each feedforward neuron in the audio encoder by contrasting its activation on the real waveform with that on a noise reference lacking the real audio’s acoustic information. IAAN then amplifies a small set of the highestscoring neurons at inference. Across ten non-semantic speech attributes, IAAN improves average accuracy by 25.7 points on Audio-Flamingo- 3, 21.4 on Qwen2.5-Omni, and 9.7 on Kimi-Audio. It also continues to improve a model that has already been explicitly fine-tuned to prioritize acoustic evidence. In controlled comparisons, both the encoder locus and neuron-level selectivity prove necessary for this gain. Intervening after the encoder, at the decoding side or inside the language model, yields little to no improvement, or even deteriorates accuracy. The improvement also depends on which specific neurons are amplified, not merely on their number, confirming that IAAN’s acoustic score succeeds in identifying the neurons that matter. These results show that a small, precisely targeted intervention inside the audio encoder is an effective and largely untapped way to strengthen the acoustic understanding of LALMs, opening a new direction for inference-time methods that improve acoustic perception through neuron-level access to the encoder.  \nIndex Terms—large audio-language models, paralinguistic understanding, audio encoder, inference-time intervention, neuron activation steering  \nI. INTRODUCTION  \nLarge audio-language models (LALMs) [1]–[13] build on a pretrained large language model (LLM) backbone with an audio encoder [14], [15], extending auditory understanding [16]–[18] to the LLM. Speech carries rich information beyond the words themselves, such as a speaker’s emotion or gender, yet current LALMs handle semantic content well while performing markedly worse on such finegrained, non-semantic attributes [19]–[27] . Our goal is to strengthen this weaker side, making LALMs perceive acoustic detail in the voice more reliably without retraining.  \nMost inference-time efforts to close this gap intervene on the language model backbone of LALMs or in its decoding process, through contrastive decoding over output logits [28]–[30] or steering of the backbone’s hidden states, attention, or neurons [16], [31]–[35] . Because these operate after the audio representations reach the backbone, their effectiveness is bounded by how much acoustic evidence those representations still carry [36], [37] . The audio encoder, where that evidence first arises [38]–[40], has received comparatively little attention as a site of intervention. A separate question is the  \n*Equal contribution.  \ngranularity at which such an intervention should act. Output logits offer only a coarse handle on the model’s overall behavior, and hidden-state interventions, th","cbCaiugQ1e9QaZrA","https://ap.wps.com/l/cbCaiugQ1e9QaZrA","pdf",717221,1,9,"English","en",105,"# Introduction\n## Problem: weak non-semantic acoustic attribute understanding\n## Prior work: interventions after the audio representations\n## Proposed method: IAAN inside the audio encoder\n## Evaluation and results on multiple LALMs","[{\"question\":\"What problem does IAAN address in large audio-language models?\",\"answer\":\"IAAN targets the gap where LALMs perform well on speech content but poorly on fine-grained non-semantic attributes like emotion and other paralinguistic cues. It aims to strengthen acoustic understanding without retraining.\"},{\"question\":\"How does IAAN score and select neurons during inference?\",\"answer\":\"IAAN scores each feedforward neuron in the audio encoder by contrasting its activation on the real waveform with that on a noise reference lacking the real acoustic information. It then amplifies only a small set of the highest-scoring neurons at inference.\"},{\"question\":\"Why is intervening inside the audio encoder more effective than later-stage interventions?\",\"answer\":\"Controlled comparisons indicate that both the encoder intervention location and neuron-level selectivity are necessary for the improvement. Intervening after the encoder—at decoding or inside the language model—yields little to no gain and may even degrade accuracy.\"}]",1784210496,23,{"code":4,"msg":30,"data":31},"ok",{"site_id":24,"language":23,"slug":32,"title":13,"keywords":33,"description":14,"schema_data":34,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":27},"encoder-side-neuron-identification-and-amplification-for-acoustic-perception-in-large-audio-language-models","",{"@graph":35,"@context":85},[36,53,68],{"@type":37,"itemListElement":38},"BreadcrumbList",[39,43,47,50],{"item":40,"name":41,"@type":42,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":44,"name":45,"@type":42,"position":46},"https://docshare.wps.com/document/","Document",2,{"item":48,"name":12,"@type":42,"position":49},"https://docshare.wps.com/document/technology/",3,{"item":51,"name":13,"@type":42,"position":52},"https://docshare.wps.com/document/encoder-side-neuron-identification-and-amplification-for-acoustic-perception-in-large-audio-language-models/86326/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":23,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":40,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-17","2026-07-16",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"What problem does IAAN address in large audio-language models?","Question",{"text":75,"@type":76},"IAAN targets the gap where LALMs perform well on speech content but poorly on fine-grained non-semantic attributes like emotion and other paralinguistic cues. It aims to strengthen acoustic understanding without retraining.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does IAAN score and select neurons during inference?",{"text":80,"@type":76},"IAAN scores each feedforward neuron in the audio encoder by contrasting its activation on the real waveform with that on a noise reference lacking the real acoustic information. It then amplifies only a small set of the highest-scoring neurons at inference.",{"name":82,"@type":73,"acceptedAnswer":83},"Why is intervening inside the audio encoder more effective than later-stage interventions?",{"text":84,"@type":76},"Controlled comparisons indicate that both the encoder intervention location and neuron-level selectivity are necessary for the improvement. Intervening after the encoder—at decoding or inside the language model—yields little to no gain and may even degrade accuracy.","https://schema.org",{"og:url":51,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":51},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":92},[93,97,101,105,110,113,118,123,127,130,134],{"id":20,"doc_module":4,"doc_module_name":45,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":45,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":45,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":111,"slug":112},50,"technology",{"id":114,"doc_module":4,"doc_module_name":45,"category_name":115,"show_sort_weight":116,"slug":117},7,"Healthcare",40,"healthcare",{"id":119,"doc_module":4,"doc_module_name":45,"category_name":120,"show_sort_weight":121,"slug":122},8,"Research & Report",30,"research-report",{"id":21,"doc_module":4,"doc_module_name":45,"category_name":124,"show_sort_weight":125,"slug":126},"Religion & Spirituality",20,"religion-spirituality",{"id":125,"doc_module":4,"doc_module_name":45,"category_name":128,"show_sort_weight":125,"slug":129},"World Cup","world-cup",{"id":131,"doc_module":4,"doc_module_name":45,"category_name":132,"show_sort_weight":131,"slug":133},10,"Lifestyle","lifestyle",{"id":135,"doc_module":4,"doc_module_name":45,"category_name":136,"show_sort_weight":106,"slug":137},19,"General","general"]