{"id":"https://openalex.org/W4408355893","doi":"https://doi.org/10.1109/icassp49660.2025.10888140","title":"Whisper in Medusa\u2019s Ear: Multi-head Efficient Decoding for Transformer-based ASR","display_name":"Whisper in Medusa\u2019s Ear: Multi-head Efficient Decoding for Transformer-based ASR","publication_year":2025,"publication_date":"2025-03-12","ids":{"openalex":"https://openalex.org/W4408355893","doi":"https://doi.org/10.1109/icassp49660.2025.10888140"},"language":"en","primary_location":{"id":"doi:10.1109/icassp49660.2025.10888140","is_oa":false,"landing_page_url":"https://doi.org/10.1109/icassp49660.2025.10888140","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":"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/A5006031844","display_name":"Yael Segal","orcid":"https://orcid.org/0000-0003-2513-0752"},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Yael Segal-Feldman","raw_affiliation_strings":["aiOla Research,Israel"],"affiliations":[{"raw_affiliation_string":"aiOla Research,Israel","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5040863261","display_name":"Aviv Shamsian","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Aviv Shamsian","raw_affiliation_strings":["aiOla Research,Israel"],"affiliations":[{"raw_affiliation_string":"aiOla Research,Israel","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5076537607","display_name":"Aviv Navon","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Aviv Navon","raw_affiliation_strings":["aiOla Research,Israel"],"affiliations":[{"raw_affiliation_string":"aiOla Research,Israel","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5072070363","display_name":"Gill Hetz","orcid":"https://orcid.org/0000-0002-6477-7797"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Gill Hetz","raw_affiliation_strings":["aiOla Research,Israel"],"affiliations":[{"raw_affiliation_string":"aiOla Research,Israel","institution_ids":[]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5008847407","display_name":"Joseph Keshet","orcid":"https://orcid.org/0000-0003-2332-5783"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Joseph Keshet","raw_affiliation_strings":["aiOla Research,Israel"],"affiliations":[{"raw_affiliation_string":"aiOla Research,Israel","institution_ids":[]}]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":5,"corresponding_author_ids":["https://openalex.org/A5006031844"],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":3.1486,"has_fulltext":false,"cited_by_count":2,"citation_normalized_percentile":{"value":0.89830204,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":95,"max":96},"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/T11447","display_name":"Blind Source Separation Techniques","score":0.9527999758720398,"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/T11447","display_name":"Blind Source Separation Techniques","score":0.9527999758720398,"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/decoding-methods","display_name":"Decoding methods","score":0.7482250928878784},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.6730873584747314},{"id":"https://openalex.org/keywords/transformer","display_name":"Transformer","score":0.5926527976989746},{"id":"https://openalex.org/keywords/head","display_name":"Head (geology)","score":0.5061590075492859},{"id":"https://openalex.org/keywords/speech-recognition","display_name":"Speech recognition","score":0.424225389957428},{"id":"https://openalex.org/keywords/electrical-engineering","display_name":"Electrical engineering","score":0.19667744636535645},{"id":"https://openalex.org/keywords/engineering","display_name":"Engineering","score":0.14951291680335999},{"id":"https://openalex.org/keywords/telecommunications","display_name":"Telecommunications","score":0.12792715430259705},{"id":"https://openalex.org/keywords/voltage","display_name":"Voltage","score":0.10255679488182068},{"id":"https://openalex.org/keywords/geology","display_name":"Geology","score":0.07134342193603516}],"concepts":[{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.7482250928878784},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6730873584747314},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.5926527976989746},{"id":"https://openalex.org/C2780312720","wikidata":"https://www.wikidata.org/wiki/Q5689100","display_name":"Head (geology)","level":2,"score":0.5061590075492859},{"id":"https://openalex.org/C28490314","wikidata":"https://www.wikidata.org/wiki/Q189436","display_name":"Speech recognition","level":1,"score":0.424225389957428},{"id":"https://openalex.org/C119599485","wikidata":"https://www.wikidata.org/wiki/Q43035","display_name":"Electrical engineering","level":1,"score":0.19667744636535645},{"id":"https://openalex.org/C127413603","wikidata":"https://www.wikidata.org/wiki/Q11023","display_name":"Engineering","level":0,"score":0.14951291680335999},{"id":"https://openalex.org/C76155785","wikidata":"https://www.wikidata.org/wiki/Q418","display_name":"Telecommunications","level":1,"score":0.12792715430259705},{"id":"https://openalex.org/C165801399","wikidata":"https://www.wikidata.org/wiki/Q25428","display_name":"Voltage","level":2,"score":0.10255679488182068},{"id":"https://openalex.org/C127313418","wikidata":"https://www.wikidata.org/wiki/Q1069","display_name":"Geology","level":0,"score":0.07134342193603516},{"id":"https://openalex.org/C114793014","wikidata":"https://www.wikidata.org/wiki/Q52109","display_name":"Geomorphology","level":1,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/icassp49660.2025.10888140","is_oa":false,"landing_page_url":"https://doi.org/10.1109/icassp49660.2025.10888140","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":[{"id":"https://metadata.un.org/sdg/7","display_name":"Affordable and clean energy","score":0.44999998807907104}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":18,"referenced_works":["https://openalex.org/W1494198834","https://openalex.org/W3119308075","https://openalex.org/W4212797134","https://openalex.org/W4391021797","https://openalex.org/W4395112660","https://openalex.org/W4402672007","https://openalex.org/W4402683901","https://openalex.org/W6751104502","https://openalex.org/W6754244489","https://openalex.org/W6847363464","https://openalex.org/W6847386241","https://openalex.org/W6849530321","https://openalex.org/W6852423491","https://openalex.org/W6855650390","https://openalex.org/W6857879404","https://openalex.org/W6860434696","https://openalex.org/W6866223206","https://openalex.org/W6866506555"],"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/W4396696052"],"abstract_inverted_index":{"Large":[0],"transformer-based":[1],"models":[2,40],"have":[3,49],"significant":[4],"potential":[5,30],"for":[6],"speech":[7],"transcription":[8],"and":[9,14,23,37,59,115],"translation.":[10],"Their":[11],"self-attention":[12],"mechanisms":[13],"parallel":[15],"processing":[16,74],"enable":[17],"them":[18],"to":[19,42,52,72],"capture":[20],"complex":[21],"patterns":[22],"dependencies":[24],"in":[25,99,103],"audio":[26],"sequences.":[27],"However,":[28],"this":[29,63],"comes":[31],"with":[32,76],"challenges,":[33],"as":[34],"these":[35],"large":[36],"computationally":[38],"intensive":[39],"lead":[41],"slow":[43],"inference":[44],"speeds.":[45],"Various":[46],"optimization":[47],"strategies":[48],"been":[50],"proposed":[51,85],"improve":[53],"performance,":[54],"including":[55],"efficient":[56],"hardware":[57],"utilization":[58],"algorithmic":[60],"enhancements.":[61],"In":[62],"paper,":[64],"we":[65],"introduce":[66],"Whisper-Medusa,":[67],"a":[68,100],"novel":[69],"approach":[70],"designed":[71],"enhance":[73],"speed":[75],"minimal":[77],"impact":[78],"on":[79],"Word":[80],"Error":[81],"Rate":[82],"(WER).":[83],"The":[84],"model":[86],"extends":[87],"the":[88,107],"OpenAI\u2019s":[89],"Whisper":[90],"architecture":[91],"by":[92],"predicting":[93],"multiple":[94],"tokens":[95],"per":[96],"iteration,":[97],"resulting":[98],"50%":[101],"reduction":[102],"latency.":[104],"We":[105],"showcase":[106],"effectiveness":[108],"of":[109],"Whisper-Medusa":[110],"across":[111],"different":[112],"learning":[113],"setups":[114],"datasets.":[116]},"counts_by_year":[{"year":2025,"cited_by_count":2}],"updated_date":"2025-12-21T23:12:01.093139","created_date":"2025-10-10T00:00:00"}