{"id":"https://openalex.org/W7157143801","doi":"https://doi.org/10.48550/arxiv.2604.23717","title":"HeadRouter: Dynamic Head-Weight Routing for Task-Adaptive Audio Token Pruning in Large Audio Language Models","display_name":"HeadRouter: Dynamic Head-Weight Routing for Task-Adaptive Audio Token Pruning in Large Audio Language Models","publication_year":2026,"publication_date":"2026-04-26","ids":{"openalex":"https://openalex.org/W7157143801","doi":"https://doi.org/10.48550/arxiv.2604.23717"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2604.23717","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.23717","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2604.23717","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5126865411","display_name":"Peize He","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"He, Peize","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5134783291","display_name":"Yaodi Luo","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Luo, Yaodi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5134751557","display_name":"Xiaoqian Liu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Xiaoqian","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5121190807","display_name":"Xuyang Liu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Xuyang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5134754300","display_name":"Jiahang Deng","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Deng, Jiahang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5134808768","display_name":"Yaosong Du","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Du, Yaosong","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5134764718","display_name":"Bangyu Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Bangyu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5123904969","display_name":"Xiyan Gui","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Gui, Xiyan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5134817598","display_name":"Yuxuan Chen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Yuxuan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5134750578","display_name":"Linfeng Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Linfeng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10860","display_name":"Speech and Audio Processing","score":0.3422999978065491,"subfield":{"id":"https://openalex.org/subfields/1711","display_name":"Signal Processing"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T10860","display_name":"Speech and Audio Processing","score":0.3422999978065491,"subfield":{"id":"https://openalex.org/subfields/1711","display_name":"Signal Processing"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T11309","display_name":"Music and Audio Processing","score":0.21170000731945038,"subfield":{"id":"https://openalex.org/subfields/1711","display_name":"Signal Processing"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10201","display_name":"Speech Recognition and Synthesis","score":0.20909999310970306,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/security-token","display_name":"Security token","score":0.7943999767303467},{"id":"https://openalex.org/keywords/pruning","display_name":"Pruning","score":0.5641000270843506},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.5152000188827515},{"id":"https://openalex.org/keywords/dynamic-range-compression","display_name":"Dynamic range compression","score":0.5074999928474426},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.4724999964237213},{"id":"https://openalex.org/keywords/data-compression","display_name":"Data compression","score":0.38029998540878296}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8210999965667725},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.7943999767303467},{"id":"https://openalex.org/C108010975","wikidata":"https://www.wikidata.org/wiki/Q500094","display_name":"Pruning","level":2,"score":0.5641000270843506},{"id":"https://openalex.org/C28490314","wikidata":"https://www.wikidata.org/wiki/Q189436","display_name":"Speech recognition","level":1,"score":0.5485000014305115},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5152000188827515},{"id":"https://openalex.org/C150178126","wikidata":"https://www.wikidata.org/wiki/Q18433212","display_name":"Dynamic range compression","level":2,"score":0.5074999928474426},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.4724999964237213},{"id":"https://openalex.org/C78548338","wikidata":"https://www.wikidata.org/wiki/Q2493","display_name":"Data compression","level":2,"score":0.38029998540878296},{"id":"https://openalex.org/C180016635","wikidata":"https://www.wikidata.org/wiki/Q2712821","display_name":"Compression (physics)","level":2,"score":0.3765000104904175},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3517000079154968},{"id":"https://openalex.org/C127220857","wikidata":"https://www.wikidata.org/wiki/Q2719318","display_name":"Audio signal processing","level":4,"score":0.3379000127315521},{"id":"https://openalex.org/C74172769","wikidata":"https://www.wikidata.org/wiki/Q1446839","display_name":"Routing (electronic design automation)","level":2,"score":0.3276999890804291},{"id":"https://openalex.org/C13895895","wikidata":"https://www.wikidata.org/wiki/Q3270773","display_name":"Speech coding","level":2,"score":0.31209999322891235},{"id":"https://openalex.org/C184337299","wikidata":"https://www.wikidata.org/wiki/Q1437428","display_name":"Semantics (computer science)","level":2,"score":0.28540000319480896},{"id":"https://openalex.org/C160372630","wikidata":"https://www.wikidata.org/wiki/Q4819855","display_name":"Audio analyzer","level":5,"score":0.26350000500679016},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.26100000739097595}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2604.23717","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.23717","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2604.23717","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.23717","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Recent":[0],"large":[1],"audio":[2,48,72,122,166],"language":[3],"models":[4],"(LALMs)":[5],"demonstrate":[6,149],"remarkable":[7],"capabilities":[8],"in":[9,30,42,120],"processing":[10],"extended":[11],"multi-modal":[12],"sequences,":[13],"yet":[14],"incur":[15],"high":[16],"inference":[17],"costs.":[18],"Token":[19],"compression":[20,34,154],"is":[21,132],"an":[22],"effective":[23],"method":[24,111],"that":[25,38,64,77,112,150],"directly":[26],"reduces":[27],"redundant":[28],"tokens":[29,167],"the":[31,114,126,144,157,165,174],"sequence.":[32],"Existing":[33],"methods":[35],"usually":[36],"assume":[37],"all":[39,58],"attention":[40,65,83,118],"heads":[41,66,84,119],"LALMs":[43],"contribute":[44],"equally":[45],"to":[46,87,124,138],"various":[47,139],"tasks":[49,123],"and":[50,96,134,146,168,171,179],"calculate":[51],"token":[52,109],"importance":[53,116],"by":[54],"averaging":[55],"scores":[56],"across":[57,70],"heads.":[59],"However,":[60],"our":[61],"analysis":[62],"demonstrates":[63],"exhibit":[67],"distinct":[68],"behaviors":[69],"diverse":[71],"domains.":[73],"We":[74],"further":[75],"reveal":[76],"only":[78],"a":[79,107],"sparse":[80],"subset":[81],"of":[82,101,117,128,164,173],"actively":[85],"responds":[86],"audio,":[88],"with":[89],"completely":[90],"different":[91,121],"performance":[92],"when":[93,161],"handling":[94],"semantic":[95],"acoustic":[97],"tasks.":[98],"In":[99],"light":[100],"this":[102],"observation,":[103],"we":[104],"propose":[105],"HeadRouter,":[106],"head-importance-aware":[108],"pruning":[110],"perceives":[113],"varying":[115],"maximize":[125],"retention":[127],"crucial":[129],"tokens.":[130],"HeadRouter":[131,151],"training-free":[133],"can":[135],"be":[136],"applied":[137],"LALMs.":[140],"Extensive":[141],"experiments":[142],"on":[143,177],"AudioMarathon":[145],"MMAU-Pro":[147],"benchmarks":[148],"achieves":[152],"state-of-the-art":[153],"performance,":[155],"exceeding":[156],"baseline":[158],"model":[159],"even":[160],"retaining":[162],"70%":[163],"achieving":[169],"101.8%":[170],"103.0%":[172],"vanilla":[175],"average":[176],"Qwen2.5-Omni-3B":[178],"Qwen2.5-Omni-7B,":[180],"respectively.":[181]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-04-29T00:00:00"}