{"id":"https://openalex.org/W6929327442","doi":"https://doi.org/10.48550/arxiv.2507.18181","title":"SpecASR: Accelerating LLM-based Automatic Speech Recognition via Speculative Decoding","display_name":"SpecASR: Accelerating LLM-based Automatic Speech Recognition via Speculative Decoding","publication_year":2025,"publication_date":"2025-07-24","ids":{"openalex":"https://openalex.org/W6929327442","doi":"https://doi.org/10.48550/arxiv.2507.18181"},"language":"en","primary_location":{"id":"doi:10.48550/arxiv.2507.18181","doi":"https://doi.org/10.48550/arxiv.2507.18181","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2507.18181","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"volume":null,"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2507.18181","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5090789582","display_name":"Wei Linye","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wei, Linye","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5081222727","display_name":"Shuzhang Zhong","orcid":"https://orcid.org/0009-0006-5478-3604","observed_orcids":["https://orcid.org/0009-0006-5478-3604"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhong, Shuzhang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5113389019","display_name":"Songqiang Xu","orcid":"https://orcid.org/0009-0001-8373-0732","observed_orcids":["https://orcid.org/0009-0001-8373-0732"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xu, Songqiang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Wang, Runsheng","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Runsheng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Huang, Ru","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Huang, Ru","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":null,"display_name":"Li, Meng","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Meng","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":true,"primary_topic":{"id":"https://openalex.org/T10201","display_name":"Speech Recognition and Synthesis","score":0.9990000128746033,"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":0.9990000128746033,"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":0.0003000000142492354,"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/T10860","display_name":"Speech and Audio Processing","score":0.00019999999494757503,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/speculative-decoding","display_name":"speculative decoding","score":0.996999979019165},{"id":"https://openalex.org/keywords/decoding-latency-reduction","display_name":"decoding latency reduction","score":0.8149999976158142}],"concepts":[{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.8787000179290771},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.8004000186920166},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.794700026512146},{"id":"https://openalex.org/C28490314","wikidata":"https://www.wikidata.org/wiki/Q189436","display_name":"Speech recognition","level":1,"score":0.6739000082015991},{"id":"https://openalex.org/C193969084","wikidata":"https://www.wikidata.org/wiki/Q7452500","display_name":"Sequential decoding","level":4,"score":0.5418000221252441},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.5414999723434448},{"id":"https://openalex.org/C2778112365","wikidata":"https://www.wikidata.org/wiki/Q3511065","display_name":"Sequence (biology)","level":2,"score":0.5056999921798706},{"id":"https://openalex.org/C82876162","wikidata":"https://www.wikidata.org/wiki/Q17096504","display_name":"Latency (audio)","level":2,"score":0.46720001101493835},{"id":"https://openalex.org/C46637626","wikidata":"https://www.wikidata.org/wiki/Q6693015","display_name":"Low latency (capital markets)","level":2,"score":0.4629000127315521},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.4302999973297119},{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.4244000017642975},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.40779998898506165},{"id":"https://openalex.org/C98045186","wikidata":"https://www.wikidata.org/wiki/Q205663","display_name":"Process (computing)","level":2,"score":0.3637999892234802},{"id":"https://openalex.org/C159877910","wikidata":"https://www.wikidata.org/wiki/Q2202883","display_name":"Autoregressive model","level":2,"score":0.32420000433921814},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.2897999882698059},{"id":"https://openalex.org/C157764524","wikidata":"https://www.wikidata.org/wiki/Q1383412","display_name":"Throughput","level":3,"score":0.2867000102996826},{"id":"https://openalex.org/C186644900","wikidata":"https://www.wikidata.org/wiki/Q194152","display_name":"Parsing","level":2,"score":0.2847999930381775},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.2825999855995178},{"id":"https://openalex.org/C113174947","wikidata":"https://www.wikidata.org/wiki/Q2859736","display_name":"Tree (set theory)","level":2,"score":0.2606000006198883},{"id":"https://openalex.org/C13895895","wikidata":"https://www.wikidata.org/wiki/Q3270773","display_name":"Speech coding","level":2,"score":0.257099986076355},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.25619998574256897},{"id":"https://openalex.org/C175154964","wikidata":"https://www.wikidata.org/wiki/Q380077","display_name":"Task analysis","level":3,"score":0.2556000053882599},{"id":"https://openalex.org/C40506919","wikidata":"https://www.wikidata.org/wiki/Q7452469","display_name":"Sequence learning","level":2,"score":0.2538999915122986}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2507.18181","doi":"https://doi.org/10.48550/arxiv.2507.18181","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2507.18181","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2507.18181","doi":"https://doi.org/10.48550/arxiv.2507.18181","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2507.18181","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[],"sustainable_development_goals_aurora":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Large":[0],"language":[1],"model":[2,158],"(LLM)-based":[3],"automatic":[4],"speech":[5],"recognition":[6,19,206],"(ASR)":[7],"has":[8,40],"recently":[9],"attracted":[10],"a":[11,73,141,161],"lot":[12],"of":[13,30,53,175],"attention":[14],"due":[15],"to":[16,132,153,171],"its":[17],"high":[18,27,99],"accuracy":[20],"and":[21,57,104,177,190,198],"enhanced":[22],"multi-dialect":[23],"support.":[24],"However,":[25],"the":[26,33,50,54,64,128,134,148,155,173,194],"decoding":[28,39,45,76,93,114,197],"latency":[29,174],"LLMs":[31],"challenges":[32],"real-time":[34,65],"ASR":[35,55,66,92,106,157,179],"requirements.":[36],"Although":[37],"speculative":[38,75,199],"been":[41],"explored":[42],"for":[43,79],"better":[44],"efficiency,":[46],"they":[47],"usually":[48],"ignore":[49],"key":[51],"characteristics":[52],"task":[56],"achieve":[58],"limited":[59],"speedup.":[60],"To":[61],"further":[62,139],"reduce":[63,154],"latency,":[67],"in":[68,98,112,205],"this":[69],"paper,":[70],"we":[71,185],"propose":[72],"novel":[74],"framework":[77],"specialized":[78],"ASR,":[80],"dubbed":[81],"SpecASR.":[82],"SpecASR":[83,117,138,187],"is":[84,94,168],"developed":[85],"based":[86],"on":[87],"our":[88],"core":[89],"observation":[90],"that":[91,125,146],"audio-conditioned,":[95],"which":[96],"results":[97],"output":[100,110],"alignment":[101],"between":[102],"small":[103],"large":[105],"models,":[107],"even":[108],"given":[109],"mismatches":[111],"intermediate":[113],"steps.":[115],"Therefore,":[116],"features":[118],"an":[119],"adaptive":[120],"draft":[121,129,142,151,156,176],"sequence":[122,130,143,152],"generation":[123,166],"process":[124],"dynamically":[126],"modifies":[127],"length":[131],"maximize":[133],"token":[135,164],"acceptance":[136],"length.":[137],"proposes":[140],"recycling":[144],"strategy":[145],"reuses":[147],"previously":[149],"generated":[150],"latency.":[159],"Moreover,":[160],"two-pass":[162],"sparse":[163],"tree":[165],"algorithm":[167],"also":[169],"proposed":[170],"balance":[172],"target":[178],"models.":[180],"With":[181],"extensive":[182],"experimental":[183],"results,":[184],"demonstrate":[186],"achieves":[188],"3.04x-3.79x":[189],"1.25x-1.84x":[191],"speedup":[192],"over":[193],"baseline":[195],"autoregressive":[196],"decoding,":[200],"respectively,":[201],"without":[202],"any":[203],"loss":[204],"accuracy.":[207]},"counts_by_year":[],"updated_date":"2026-10-14T16:39:01.820549","created_date":"2025-10-10T00:00:00"}