{"id":"https://openalex.org/W7166859430","doi":"https://doi.org/10.18653/v1/2026.findings-acl.742","title":"Do We Need Distinct Representations for Every Speech Token? Unveiling and Exploiting Redundancy in Large Speech Language Models","display_name":"Do We Need Distinct Representations for Every Speech Token? Unveiling and Exploiting Redundancy in Large Speech Language Models","publication_year":2026,"publication_date":"2026-01-01","ids":{"openalex":"https://openalex.org/W7166859430","doi":"https://doi.org/10.18653/v1/2026.findings-acl.742"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2026.findings-acl.742","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.742","pdf_url":"https://aclanthology.org/2026.findings-acl.742.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2026","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2026.findings-acl.742.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5068783537","display_name":"Bajian Xiang","orcid":"https://orcid.org/0009-0005-5151-2115"},"institutions":[{"id":"https://openalex.org/I4210128412","display_name":"Beike Biotechnology (China)","ror":"https://ror.org/03fd44j44","country_code":"CN","type":"company","lineage":["https://openalex.org/I4210128412"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Bajian Xiang","raw_affiliation_strings":["Beike Inc. , Beijing , China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Beike Inc. , Beijing , China","institution_ids":["https://openalex.org/I4210128412"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139815177","display_name":"Tingwei Guo","orcid":null},"institutions":[{"id":"https://openalex.org/I4210128412","display_name":"Beike Biotechnology (China)","ror":"https://ror.org/03fd44j44","country_code":"CN","type":"company","lineage":["https://openalex.org/I4210128412"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Tingwei Guo","raw_affiliation_strings":["Beike Inc. , Beijing , China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Beike Inc. , Beijing , China","institution_ids":["https://openalex.org/I4210128412"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139757779","display_name":"Xuan Chen","orcid":null},"institutions":[{"id":"https://openalex.org/I4210128412","display_name":"Beike Biotechnology (China)","ror":"https://ror.org/03fd44j44","country_code":"CN","type":"company","lineage":["https://openalex.org/I4210128412"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Xuan Chen","raw_affiliation_strings":["Beike Inc. , Beijing , China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Beike Inc. , Beijing , China","institution_ids":["https://openalex.org/I4210128412"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5139758352","display_name":"Han Yang","orcid":null},"institutions":[{"id":"https://openalex.org/I4210128412","display_name":"Beike Biotechnology (China)","ror":"https://ror.org/03fd44j44","country_code":"CN","type":"company","lineage":["https://openalex.org/I4210128412"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yang Han","raw_affiliation_strings":["Beike Inc. , Beijing , China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Beike Inc. , Beijing , China","institution_ids":["https://openalex.org/I4210128412"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I4210128412"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.84809554,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"15069","last_page":"15087"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10201","display_name":"Speech Recognition and Synthesis","score":0.4417000114917755,"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"}},"topics":[{"id":"https://openalex.org/T10201","display_name":"Speech Recognition and Synthesis","score":0.4417000114917755,"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"}},{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.1467999964952469,"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"}},{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.0966000035405159,"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/redundancy","display_name":"Redundancy (engineering)","score":0.5376999974250793},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.40639999508857727},{"id":"https://openalex.org/keywords/natural-language","display_name":"Natural language","score":0.38350000977516174},{"id":"https://openalex.org/keywords/speech-processing","display_name":"Speech processing","score":0.31150001287460327},{"id":"https://openalex.org/keywords/focus","display_name":"Focus (optics)","score":0.27230000495910645}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7093999981880188},{"id":"https://openalex.org/C152124472","wikidata":"https://www.wikidata.org/wiki/Q1204361","display_name":"Redundancy (engineering)","level":2,"score":0.5376999974250793},{"id":"https://openalex.org/C28490314","wikidata":"https://www.wikidata.org/wiki/Q189436","display_name":"Speech recognition","level":1,"score":0.5216000080108643},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.5199000239372253},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.47200000286102295},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.40639999508857727},{"id":"https://openalex.org/C195324797","wikidata":"https://www.wikidata.org/wiki/Q33742","display_name":"Natural language","level":2,"score":0.38350000977516174},{"id":"https://openalex.org/C61328038","wikidata":"https://www.wikidata.org/wiki/Q3358061","display_name":"Speech processing","level":2,"score":0.31150001287460327},{"id":"https://openalex.org/C192209626","wikidata":"https://www.wikidata.org/wiki/Q190909","display_name":"Focus (optics)","level":2,"score":0.27230000495910645},{"id":"https://openalex.org/C129792486","wikidata":"https://www.wikidata.org/wiki/Q1050419","display_name":"Language identification","level":3,"score":0.24500000476837158}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2026.findings-acl.742","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.742","pdf_url":"https://aclanthology.org/2026.findings-acl.742.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2026","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2026.findings-acl.742","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.742","pdf_url":"https://aclanthology.org/2026.findings-acl.742.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2026","raw_type":"proceedings-article"},"sustainable_development_goals":[{"id":"https://metadata.un.org/sdg/10","display_name":"Reduced inequalities","score":0.4798261225223541}],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7166859430.pdf","grobid_xml":"https://content.openalex.org/works/W7166859430.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Large":[0],"Speech":[1],"Language":[2],"Models":[3],"(LSLMs)":[4],"typically":[5],"operate":[6],"at":[7,87],"high":[8],"token":[9,80,145],"rates":[10],"(tokens/s)":[11],"to":[12,127],"ensure":[13],"acoustic":[14,59],"fidelity,":[15],"yet":[16],"this":[17,33,85],"results":[18,138],"in":[19],"sequence":[20],"lengths":[21],"that":[22,107],"far":[23],"exceed":[24],"the":[25,38,140],"underlying":[26],"semantic":[27,100],"content,":[28],"incurring":[29],"prohibitive":[30],"inference":[31],"costs.In":[32],"paper,":[34],"we":[35,48,73,93],"empirically":[36],"revisit":[37],"necessity":[39,141],"of":[40,142],"such":[41],"granular":[42],"token-level":[43],"processing.Through":[44],"layer-wise":[45],"oracle":[46],"interventions,":[47],"unveil":[49],"a":[50,77],"structured":[51],"redundancy":[52],"hierarchy:":[53],"while":[54,115],"shallow":[55],"layers":[56,62],"encode":[57],"essential":[58],"details,":[60],"deep":[61,91],"exhibit":[63],"extreme":[64],"redundancy,":[65],"allowing":[66],"for":[67],"aggressive":[68],"compression.Motivated":[69],"by":[70,113],"these":[71],"findings,":[72],"introduce":[74],"Affinity":[75],"Pooling,":[76],"training-free,":[78],"similarity-based":[79],"merging":[81],"mechanism.By":[82],"strategically":[83],"applying":[84],"method":[86],"both":[88],"input":[89],"and":[90,131],"layers,":[92],"effectively":[94],"compress":[95],"speech":[96],"representations":[97],"without":[98],"compromising":[99],"information.Extensive":[101],"evaluations":[102],"across":[103],"three":[104],"tasks":[105],"demonstrate":[106],"our":[108],"approach":[109],"reduces":[110],"prefilling":[111],"FLOPs":[112],"27.48%":[114],"maintaining":[116],"competitive":[117],"accuracy.Practical":[118],"deployment":[119],"further":[120],"confirms":[121],"significant":[122],"efficiency":[123],"gains,":[124],"yielding":[125],"up":[126],"\u223c1.7\u00d7":[128],"memory":[129],"savings":[130],"\u223c1.1\u00d7":[132],"faster":[133],"time-to-first-token":[134],"on":[135,150],"long":[136],"utterances.Our":[137],"challenge":[139],"fully":[143],"distinct":[144],"representations,":[146],"providing":[147],"new":[148],"perspectives":[149],"LSLM":[151],"efficiency.":[152]},"counts_by_year":[],"updated_date":"2026-07-29T14:22:42.915294","created_date":"2026-07-02T00:00:00"}