{"id":"https://openalex.org/W4408353886","doi":"https://doi.org/10.1109/icassp49660.2025.10890533","title":"UME: Upcycling Mixture-of-Experts for Scalable and Efficient Automatic Speech Recognition","display_name":"UME: Upcycling Mixture-of-Experts for Scalable and Efficient Automatic Speech Recognition","publication_year":2025,"publication_date":"2025-03-12","ids":{"openalex":"https://openalex.org/W4408353886","doi":"https://doi.org/10.1109/icassp49660.2025.10890533"},"language":"en","primary_location":{"id":"doi:10.1109/icassp49660.2025.10890533","is_oa":false,"landing_page_url":"https://doi.org/10.1109/icassp49660.2025.10890533","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"ICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":false,"oa_status":"closed","oa_url":null,"any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5101865288","display_name":"Li Fu","orcid":"https://orcid.org/0000-0002-7533-692X"},"institutions":[{"id":"https://openalex.org/I4210103986","display_name":"Jingdong (China)","ror":"https://ror.org/01dkjkq64","country_code":"CN","type":"company","lineage":["https://openalex.org/I4210103986"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Li Fu","raw_affiliation_strings":["JD AI Research,Beijing,China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"JD AI Research,Beijing,China","institution_ids":["https://openalex.org/I4210103986"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5054984909","display_name":"Shanyong Yu","orcid":null},"institutions":[{"id":"https://openalex.org/I4210103986","display_name":"Jingdong (China)","ror":"https://ror.org/01dkjkq64","country_code":"CN","type":"company","lineage":["https://openalex.org/I4210103986"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Shanyong Yu","raw_affiliation_strings":["JD AI Research,Beijing,China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"JD AI Research,Beijing,China","institution_ids":["https://openalex.org/I4210103986"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100336560","display_name":"Siqi Li","orcid":"https://orcid.org/0000-0003-2085-4938"},"institutions":[{"id":"https://openalex.org/I4210103986","display_name":"Jingdong (China)","ror":"https://ror.org/01dkjkq64","country_code":"CN","type":"company","lineage":["https://openalex.org/I4210103986"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Siqi Li","raw_affiliation_strings":["JD AI Research,Beijing,China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"JD AI Research,Beijing,China","institution_ids":["https://openalex.org/I4210103986"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100371061","display_name":"Lu Fan","orcid":"https://orcid.org/0009-0002-9253-0019"},"institutions":[{"id":"https://openalex.org/I4210103986","display_name":"Jingdong (China)","ror":"https://ror.org/01dkjkq64","country_code":"CN","type":"company","lineage":["https://openalex.org/I4210103986"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Lu Fan","raw_affiliation_strings":["JD AI Research,Beijing,China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"JD AI Research,Beijing,China","institution_ids":["https://openalex.org/I4210103986"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5011555779","display_name":"Youzheng Wu","orcid":null},"institutions":[{"id":"https://openalex.org/I4210103986","display_name":"Jingdong (China)","ror":"https://ror.org/01dkjkq64","country_code":"CN","type":"company","lineage":["https://openalex.org/I4210103986"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Youzheng Wu","raw_affiliation_strings":["JD AI Research,Beijing,China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"JD AI Research,Beijing,China","institution_ids":["https://openalex.org/I4210103986"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5006367043","display_name":"Xiaodong He","orcid":"https://orcid.org/0000-0002-6682-2926"},"institutions":[{"id":"https://openalex.org/I4210103986","display_name":"Jingdong (China)","ror":"https://ror.org/01dkjkq64","country_code":"CN","type":"company","lineage":["https://openalex.org/I4210103986"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Xiaodong He","raw_affiliation_strings":["JD AI Research,Beijing,China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"JD AI Research,Beijing,China","institution_ids":["https://openalex.org/I4210103986"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I4210103986"],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"cited_by_count":1,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"1","last_page":"5"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10201","display_name":"Speech Recognition and Synthesis","score":0.9902999997138977,"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":0.9902999997138977,"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/T11309","display_name":"Music and Audio Processing","score":0.9516000151634216,"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/T10860","display_name":"Speech and Audio Processing","score":0.9472000002861023,"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/computer-science","display_name":"Computer science","score":0.7698105573654175},{"id":"https://openalex.org/keywords/speech-recognition","display_name":"Speech recognition","score":0.706507682800293},{"id":"https://openalex.org/keywords/scalability","display_name":"Scalability","score":0.6495386362075806},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.36781784892082214},{"id":"https://openalex.org/keywords/natural-language-processing","display_name":"Natural language processing","score":0.33721762895584106},{"id":"https://openalex.org/keywords/operating-system","display_name":"Operating system","score":0.12592685222625732}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7698105573654175},{"id":"https://openalex.org/C28490314","wikidata":"https://www.wikidata.org/wiki/Q189436","display_name":"Speech recognition","level":1,"score":0.706507682800293},{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.6495386362075806},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.36781784892082214},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.33721762895584106},{"id":"https://openalex.org/C111919701","wikidata":"https://www.wikidata.org/wiki/Q9135","display_name":"Operating system","level":1,"score":0.12592685222625732}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/icassp49660.2025.10890533","is_oa":false,"landing_page_url":"https://doi.org/10.1109/icassp49660.2025.10890533","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"ICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)","raw_type":"proceedings-article"}],"best_oa_location":null,"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":41,"referenced_works":["https://openalex.org/W2127141656","https://openalex.org/W3023953056","https://openalex.org/W3097777922","https://openalex.org/W3165404421","https://openalex.org/W3196117288","https://openalex.org/W3196950043","https://openalex.org/W3198694222","https://openalex.org/W3201225328","https://openalex.org/W3203407300","https://openalex.org/W3204696009","https://openalex.org/W3209984917","https://openalex.org/W3211278025","https://openalex.org/W3215427016","https://openalex.org/W4225295099","https://openalex.org/W4283067311","https://openalex.org/W4319587035","https://openalex.org/W4372349677","https://openalex.org/W4385567093","https://openalex.org/W4385822407","https://openalex.org/W4385822875","https://openalex.org/W4385823078","https://openalex.org/W4388017359","https://openalex.org/W4391021741","https://openalex.org/W4392679398","https://openalex.org/W4392904112","https://openalex.org/W4392909818","https://openalex.org/W4402672077","https://openalex.org/W4403318601","https://openalex.org/W6685444988","https://openalex.org/W6732520560","https://openalex.org/W6739901393","https://openalex.org/W6784614252","https://openalex.org/W6788811087","https://openalex.org/W6796581206","https://openalex.org/W6847237767","https://openalex.org/W6847363464","https://openalex.org/W6850218400","https://openalex.org/W6860710830","https://openalex.org/W6861293806","https://openalex.org/W6865554915","https://openalex.org/W6869986516"],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2899084033","https://openalex.org/W2748952813","https://openalex.org/W2390279801","https://openalex.org/W4391913857","https://openalex.org/W2358668433","https://openalex.org/W4396701345","https://openalex.org/W2376932109","https://openalex.org/W2001405890","https://openalex.org/W3204019825"],"abstract_inverted_index":{"Recent":[0],"advancements":[1],"in":[2,10],"scaling":[3],"up":[4,126],"models":[5,20,136],"have":[6],"significantly":[7,69],"improved":[8],"performance":[9],"Automatic":[11],"Speech":[12],"Recognition":[13],"(ASR)":[14],"tasks.":[15],"However,":[16],"training":[17,84,123,135],"large":[18],"ASR":[19,40],"from":[21,141],"scratch":[22],"remains":[23],"costly.":[24],"To":[25],"address":[26],"this":[27],"issue,":[28],"we":[29,60],"introduce":[30],"UME,":[31],"a":[32,62,92,109],"novel":[33],"method":[34],"that":[35,101],"efficiently":[36],"Upcycles":[37],"pretrained":[38,58,106],"dense":[39],"checkpoints":[41],"into":[42,52],"larger":[43],"Mixture-of-Eperts":[44],"(MoE)":[45],"architectures.":[46],"Initially,":[47],"feed-forward":[48],"networks":[49],"are":[50,80],"converted":[51],"MoE":[53],"layers.":[54],"By":[55],"reusing":[56],"the":[57,66,85,105,138],"weights,":[59],"establish":[61],"robust":[63],"foundation":[64],"for":[65],"expanded":[67],"model,":[68,86],"reducing":[70],"optimization":[71],"time.":[72],"Then,":[73],"layer":[74],"freezing":[75],"and":[76,97,129],"expert":[77],"balancing":[78],"strategies":[79],"employed":[81],"to":[82,127,134],"continue":[83],"further":[87],"enhancing":[88],"performance.":[89],"Experiments":[90],"on":[91],"mixture":[93],"of":[94,111,137],"170k-hour":[95],"Mandarin":[96],"English":[98],"datasets":[99],"show":[100],"UME:":[102],"1)":[103],"surpasses":[104],"baseline":[107],"by":[108,125],"margin":[110],"11.9%":[112],"relative":[113],"error":[114],"rate":[115],"reduction":[116],"while":[117],"maintaining":[118],"comparable":[119],"latency;":[120],"2)":[121],"reduces":[122],"time":[124],"86.7%":[128],"achieves":[130],"superior":[131],"accuracy":[132],"compared":[133],"same":[139],"size":[140],"scratch.":[142]},"counts_by_year":[{"year":2025,"cited_by_count":1}],"updated_date":"2026-07-14T23:27:15.235271","created_date":"2025-10-10T00:00:00"}