{"id":"https://openalex.org/W4392903440","doi":"https://doi.org/10.1109/icassp48485.2024.10448025","title":"TODM: Train Once Deploy Many Efficient Supernet-Based RNN-T Compression For On-Device ASR Models","display_name":"TODM: Train Once Deploy Many Efficient Supernet-Based RNN-T Compression For On-Device ASR Models","publication_year":2024,"publication_date":"2024-03-18","ids":{"openalex":"https://openalex.org/W4392903440","doi":"https://doi.org/10.1109/icassp48485.2024.10448025"},"language":"en","primary_location":{"id":"doi:10.1109/icassp48485.2024.10448025","is_oa":false,"landing_page_url":"https://doi.org/10.1109/icassp48485.2024.10448025","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/A5047358828","display_name":"Yuan Shangguan","orcid":null},"institutions":[],"countries":[],"is_corresponding":true,"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/A5069376593","display_name":"Haichuan Yang","orcid":"https://orcid.org/0000-0001-7100-7945"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Haichuan Yang","raw_affiliation_strings":["Meta AI"],"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100664691","display_name":"Danni Li","orcid":"https://orcid.org/0009-0006-0603-8562"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Danni Li","raw_affiliation_strings":["Meta AI"],"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5103012144","display_name":"Chunyang Wu","orcid":"https://orcid.org/0000-0002-5796-8288"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chunyang Wu","raw_affiliation_strings":["Meta AI"],"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5006814826","display_name":"Yassir Fathullah","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yassir Fathullah","raw_affiliation_strings":["Meta AI"],"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5016340673","display_name":"Dilin Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Dilin Wang","raw_affiliation_strings":["Meta AI"],"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5026318930","display_name":"Ayushi Dalmia","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ayushi Dalmia","raw_affiliation_strings":["Meta AI"],"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5039488725","display_name":"Raghuraman Krishnamoorthi","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Raghuraman Krishnamoorthi","raw_affiliation_strings":["Meta AI"],"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5066166549","display_name":"Ozlem Kalinli","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ozlem Kalinli","raw_affiliation_strings":["Meta AI"],"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5113970008","display_name":"Junteng Jia","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Junteng Jia","raw_affiliation_strings":["Meta AI"],"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5074237839","display_name":"Jay Mahadeokar","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jay Mahadeokar","raw_affiliation_strings":["Meta AI"],"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5106234191","display_name":"Xin Lei","orcid":"https://orcid.org/0009-0009-2028-717X"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xin Lei","raw_affiliation_strings":["Meta AI"],"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5113773386","display_name":"Mike Seltzer","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Mike Seltzer","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":14,"corresponding_author_ids":["https://openalex.org/A5047358828"],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":0.6738,"has_fulltext":false,"cited_by_count":2,"citation_normalized_percentile":{"value":0.72380462,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":91,"max":98},"biblio":{"volume":null,"issue":null,"first_page":"10216","last_page":"10220"},"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":0.9983000159263611,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.9979000091552734,"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/computer-science","display_name":"Computer science","score":0.845441997051239},{"id":"https://openalex.org/keywords/hyperparameter","display_name":"Hyperparameter","score":0.6963748931884766},{"id":"https://openalex.org/keywords/dropout","display_name":"Dropout (neural networks)","score":0.6963425874710083},{"id":"https://openalex.org/keywords/recurrent-neural-network","display_name":"Recurrent neural network","score":0.6480916738510132},{"id":"https://openalex.org/keywords/task","display_name":"Task (project management)","score":0.44656258821487427},{"id":"https://openalex.org/keywords/artificial-neural-network","display_name":"Artificial neural network","score":0.4021405577659607},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.37175652384757996},{"id":"https://openalex.org/keywords/machine-learning","display_name":"Machine learning","score":0.36296147108078003}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.845441997051239},{"id":"https://openalex.org/C8642999","wikidata":"https://www.wikidata.org/wiki/Q4171168","display_name":"Hyperparameter","level":2,"score":0.6963748931884766},{"id":"https://openalex.org/C2776145597","wikidata":"https://www.wikidata.org/wiki/Q25339462","display_name":"Dropout (neural networks)","level":2,"score":0.6963425874710083},{"id":"https://openalex.org/C147168706","wikidata":"https://www.wikidata.org/wiki/Q1457734","display_name":"Recurrent neural network","level":3,"score":0.6480916738510132},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.44656258821487427},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.4021405577659607},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.37175652384757996},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.36296147108078003},{"id":"https://openalex.org/C187736073","wikidata":"https://www.wikidata.org/wiki/Q2920921","display_name":"Management","level":1,"score":0.0},{"id":"https://openalex.org/C162324750","wikidata":"https://www.wikidata.org/wiki/Q8134","display_name":"Economics","level":0,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/icassp48485.2024.10448025","is_oa":false,"landing_page_url":"https://doi.org/10.1109/icassp48485.2024.10448025","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":[{"id":"https://metadata.un.org/sdg/7","display_name":"Affordable and clean energy","score":0.47999998927116394}],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":29,"referenced_works":["https://openalex.org/W398859631","https://openalex.org/W569478347","https://openalex.org/W1494198834","https://openalex.org/W2143612262","https://openalex.org/W2889129739","https://openalex.org/W2963250244","https://openalex.org/W2963376890","https://openalex.org/W2981698279","https://openalex.org/W3015194534","https://openalex.org/W3109946440","https://openalex.org/W3144557079","https://openalex.org/W3160766462","https://openalex.org/W3162649911","https://openalex.org/W3163203022","https://openalex.org/W3196784225","https://openalex.org/W3207326706","https://openalex.org/W4223988178","https://openalex.org/W4319862425","https://openalex.org/W4372348492","https://openalex.org/W4385822815","https://openalex.org/W4387799863","https://openalex.org/W4388017359","https://openalex.org/W6638749077","https://openalex.org/W6675365184","https://openalex.org/W6757036269","https://openalex.org/W6767064347","https://openalex.org/W6768205276","https://openalex.org/W6791110730","https://openalex.org/W6857062747"],"related_works":["https://openalex.org/W4390421286","https://openalex.org/W4280563792","https://openalex.org/W2140186469","https://openalex.org/W4389724018","https://openalex.org/W4318719684","https://openalex.org/W4318559728","https://openalex.org/W2775233965","https://openalex.org/W4360995913","https://openalex.org/W3082178636","https://openalex.org/W4312193868"],"abstract_inverted_index":{"Automatic":[0],"Speech":[1],"Recognition":[2],"(ASR)":[3],"models":[4,37,67,92,113,181,203],"need":[5],"to":[6,58,71,107,127,184],"be":[7,15,21,43],"optimized":[8],"for":[9,115],"specific":[10],"hardware":[11,117],"before":[12],"they":[13],"can":[14,20,42],"deployed":[16],"on":[17,84],"devices.":[18],"This":[19,47],"done":[22],"by":[23,152,182],"tuning":[24],"the":[25,105,129,132,143,176,198],"model\u2019s":[26],"hyperparameters":[27],"or":[28,174],"exploring":[29],"variations":[30],"in":[31,190],"its":[32],"architecture.":[33],"Re-training":[34],"and":[35,102,142],"re-validating":[36],"after":[38],"making":[39,110],"these":[40],"changes":[41],"a":[44,55,74,96,121,185,205],"resource-intensive":[45],"task.":[46],"paper":[48],"presents":[49],"TODM":[50,78,133,170],"(Train":[51],"Once":[52],"Deploy":[53],"Many),":[54],"new":[56],"approach":[57,151],"efficiently":[59,196],"train":[60],"many":[61,202],"sizes":[62,101],"of":[63,73,104,124,131,145,178,187,200],"hardware-friendly":[64],"on-device":[65],"ASR":[66],"with":[68],"comparable":[69],"GPU-hours":[70],"that":[72,168],"single":[75],"training":[76,201],"job.":[77],"leverages":[79],"insights":[80],"from":[81],"prior":[82],"work":[83],"Supernet,":[85],"where":[86],"Recurrent":[87],"Neural":[88],"Network":[89],"Transducer":[90],"(RNN-T)":[91],"share":[93],"weights":[94],"within":[95],"Supernet.":[97],"It":[98],"reduces":[99],"layer":[100],"widths":[103],"Supernet":[106,171],"obtain":[108],"subnetworks,":[109],"them":[111],"smaller":[112],"suitable":[114],"all":[116],"types.":[118],"We":[119,148],"introduce":[120],"novel":[122],"combination":[123],"three":[125],"techniques":[126],"improve":[128],"outcomes":[130],"Supernet:":[134],"adaptive":[135],"dropout,":[136],"an":[137],"in-place":[138],"Alpha-divergence":[139],"knowledge":[140],"distillation,":[141],"use":[144],"ScaledAdam":[146],"optimizer.":[147],"validate":[149],"our":[150,169],"comparing":[153],"Supernet-trained":[154],"versus":[155],"individually":[156],"tuned":[157,180],"Multi-Head":[158],"State":[159],"Space":[160],"Model":[161],"(MH-SSM)":[162],"RNN-T":[163],"using":[164],"LibriSpeech.":[165],"Results":[166],"demonstrate":[167],"either":[172],"matches":[173],"surpasses":[175],"performance":[177],"manually":[179],"up":[183],"relative":[186],"3%":[188],"better":[189],"word":[191],"error":[192],"rate":[193],"(WER),":[194],"while":[195],"keeping":[197],"cost":[199],"at":[204],"small":[206],"constant.":[207]},"counts_by_year":[{"year":2026,"cited_by_count":1},{"year":2025,"cited_by_count":1}],"updated_date":"2026-04-16T08:26:57.006410","created_date":"2025-10-10T00:00:00"}