{"id":"https://openalex.org/W4392903352","doi":"https://doi.org/10.1109/icassp48485.2024.10447827","title":"Folding Attention: Memory and Power Optimization for On-Device Transformer-Based Streaming Speech Recognition","display_name":"Folding Attention: Memory and Power Optimization for On-Device Transformer-Based Streaming Speech Recognition","publication_year":2024,"publication_date":"2024-03-18","ids":{"openalex":"https://openalex.org/W4392903352","doi":"https://doi.org/10.1109/icassp48485.2024.10447827"},"language":"en","primary_location":{"id":"doi:10.1109/icassp48485.2024.10447827","is_oa":false,"landing_page_url":"https://doi.org/10.1109/icassp48485.2024.10447827","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"ICASSP 2024 - 2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)","raw_type":"proceedings-article"},"type":"article","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/A5100421849","display_name":"Yang Li","orcid":"https://orcid.org/0009-0006-4018-5582"},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Yang Li","raw_affiliation_strings":["Meta AI"],"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5103115149","display_name":"Liangzhen Lai","orcid":"https://orcid.org/0000-0001-6421-5164"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liangzhen Lai","raw_affiliation_strings":["Meta AI"],"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5047358828","display_name":"Yuan Shangguan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yuan Shangguan","raw_affiliation_strings":["Meta AI"],"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5031485831","display_name":"Forrest Iandola","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Forrest N. Iandola","raw_affiliation_strings":["Meta AI"],"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5031088292","display_name":"Zhaoheng Ni","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhaoheng Ni","raw_affiliation_strings":["Meta AI"],"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5021305048","display_name":"Ernie Chang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ernie Chang","raw_affiliation_strings":["Meta AI"],"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5103247973","display_name":"Yangyang Shi","orcid":"https://orcid.org/0000-0001-5297-4155"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yangyang Shi","raw_affiliation_strings":["Meta AI"],"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":[]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5016704219","display_name":"Vikas Chandra","orcid":"https://orcid.org/0009-0005-4996-8455"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Vikas Chandra","raw_affiliation_strings":["Meta AI"],"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":[]}]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":8,"corresponding_author_ids":["https://openalex.org/A5100421849"],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":1.8131,"has_fulltext":false,"cited_by_count":5,"citation_normalized_percentile":{"value":0.86315547,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":94,"max":97},"biblio":{"volume":null,"issue":null,"first_page":"11901","last_page":"11905"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10201","display_name":"Speech Recognition and Synthesis","score":1.0,"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":1.0,"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":0.9997000098228455,"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":0.9995999932289124,"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/bottleneck","display_name":"Bottleneck","score":0.8143147230148315},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.809731125831604},{"id":"https://openalex.org/keywords/transformer","display_name":"Transformer","score":0.6411705017089844},{"id":"https://openalex.org/keywords/computation","display_name":"Computation","score":0.589661180973053},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.5514057874679565},{"id":"https://openalex.org/keywords/speech-recognition","display_name":"Speech recognition","score":0.47545143961906433},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.3885151743888855},{"id":"https://openalex.org/keywords/machine-learning","display_name":"Machine learning","score":0.3327710032463074},{"id":"https://openalex.org/keywords/algorithm","display_name":"Algorithm","score":0.20269709825515747},{"id":"https://openalex.org/keywords/embedded-system","display_name":"Embedded system","score":0.14966896176338196},{"id":"https://openalex.org/keywords/engineering","display_name":"Engineering","score":0.08951842784881592}],"concepts":[{"id":"https://openalex.org/C2780513914","wikidata":"https://www.wikidata.org/wiki/Q18210350","display_name":"Bottleneck","level":2,"score":0.8143147230148315},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.809731125831604},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.6411705017089844},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.589661180973053},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5514057874679565},{"id":"https://openalex.org/C28490314","wikidata":"https://www.wikidata.org/wiki/Q189436","display_name":"Speech recognition","level":1,"score":0.47545143961906433},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3885151743888855},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3327710032463074},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.20269709825515747},{"id":"https://openalex.org/C149635348","wikidata":"https://www.wikidata.org/wiki/Q193040","display_name":"Embedded system","level":1,"score":0.14966896176338196},{"id":"https://openalex.org/C127413603","wikidata":"https://www.wikidata.org/wiki/Q11023","display_name":"Engineering","level":0,"score":0.08951842784881592},{"id":"https://openalex.org/C119599485","wikidata":"https://www.wikidata.org/wiki/Q43035","display_name":"Electrical engineering","level":1,"score":0.0},{"id":"https://openalex.org/C165801399","wikidata":"https://www.wikidata.org/wiki/Q25428","display_name":"Voltage","level":2,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/icassp48485.2024.10447827","is_oa":false,"landing_page_url":"https://doi.org/10.1109/icassp48485.2024.10447827","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"ICASSP 2024 - 2024 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":36,"referenced_works":["https://openalex.org/W1494198834","https://openalex.org/W2407080277","https://openalex.org/W2531409750","https://openalex.org/W2794141774","https://openalex.org/W2802023636","https://openalex.org/W2892009249","https://openalex.org/W2911291251","https://openalex.org/W2913718171","https://openalex.org/W2946090344","https://openalex.org/W2962778134","https://openalex.org/W2963250244","https://openalex.org/W2964089206","https://openalex.org/W2972818416","https://openalex.org/W2981857663","https://openalex.org/W3007328579","https://openalex.org/W3015974384","https://openalex.org/W3033529678","https://openalex.org/W3096702180","https://openalex.org/W3097558625","https://openalex.org/W3097777922","https://openalex.org/W3106298483","https://openalex.org/W3162665866","https://openalex.org/W3163725856","https://openalex.org/W4288028629","https://openalex.org/W4288088457","https://openalex.org/W4323654151","https://openalex.org/W4372260137","https://openalex.org/W4384648484","https://openalex.org/W4385245566","https://openalex.org/W6739901393","https://openalex.org/W6761628794","https://openalex.org/W6769806307","https://openalex.org/W6770820644","https://openalex.org/W6779163297","https://openalex.org/W6796196852","https://openalex.org/W6855009501"],"related_works":["https://openalex.org/W2595172197","https://openalex.org/W2084856301","https://openalex.org/W2127970246","https://openalex.org/W2885125400","https://openalex.org/W1989889224","https://openalex.org/W4382618745","https://openalex.org/W4206178588","https://openalex.org/W3094491777","https://openalex.org/W3214715529","https://openalex.org/W4287635093"],"abstract_inverted_index":{"Transformer-based":[0,102],"models":[1,26,106],"excel":[2],"in":[3,48],"speech":[4,24,104],"recognition.":[5],"Existing":[6],"efforts":[7],"to":[8,70,120,127],"optimize":[9],"Transformer":[10],"inference,":[11],"typically":[12],"for":[13],"long-context":[14],"applications,":[15],"center":[16],"on":[17,100],"simplifying":[18],"attention":[19,37,55,110],"score":[20,38],"calculations.":[21],"However,":[22],"streaming":[23,103],"recognition":[25,105],"usually":[27],"process":[28],"a":[29,42,60,83],"limited":[30],"number":[31],"of":[32,41,53,63],"tokens":[33],"each":[34],"time,":[35],"making":[36],"calculation":[39],"less":[40],"bottleneck.":[43],"Instead,":[44],"the":[45,49,64],"bottleneck":[46],"lies":[47],"linear":[50,87],"projection":[51],"layers":[52],"multi-head":[54],"and":[56,67,73,93,96,122],"feedforward":[57],"networks,":[58],"constituting":[59],"substantial":[61],"portion":[62],"model":[65,91,112,132],"size":[66,92,113],"contributing":[68],"significantly":[69,89],"computation,":[71],"memory,":[72],"power":[74,97,123],"usage.To":[75],"address":[76],"this":[77],"bottleneck,":[78],"we":[79],"propose":[80],"folding":[81,109],"attention,":[82],"technique":[84],"targeting":[85],"these":[86],"layers,":[88],"reducing":[90],"improving":[94],"memory":[95,116],"efficiency.":[98],"Experiments":[99],"on-device":[101],"show":[107],"that":[108],"reduces":[111],"(and":[114],"corresponding":[115],"consumption)":[117],"by":[118,125],"up":[119,126],"24%":[121],"consumption":[124],"23%,":[128],"all":[129],"without":[130],"compromising":[131],"accuracy":[133],"or":[134],"computation":[135],"overhead.":[136]},"counts_by_year":[{"year":2025,"cited_by_count":3},{"year":2024,"cited_by_count":2}],"updated_date":"2025-11-06T03:46:38.306776","created_date":"2025-10-10T00:00:00"}