{"id":"https://openalex.org/W3048107166","doi":"https://doi.org/10.48550/arxiv.2008.04245","title":"TinySpeech: Attention Condensers for Deep Speech Recognition Neural Networks on Edge Devices","display_name":"TinySpeech: Attention Condensers for Deep Speech Recognition Neural Networks on Edge Devices","publication_year":2020,"publication_date":"2020-08-10","ids":{"openalex":"https://openalex.org/W3048107166","doi":"https://doi.org/10.48550/arxiv.2008.04245","mag":"3048107166"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2008.04245","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2008.04245","pdf_url":"https://arxiv.org/pdf/2008.04245","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2008.04245","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5102920751","display_name":"Alexander K.C. Wong","orcid":"https://orcid.org/0000-0001-5729-5899","observed_orcids":["https://orcid.org/0000-0001-5729-5899"]},"institutions":[{"id":"https://openalex.org/I151746483","display_name":"University of Waterloo","ror":"https://ror.org/01aff2v68","country_code":"CA","type":"education","lineage":["https://openalex.org/I151746483"]}],"countries":["CA"],"is_corresponding":false,"raw_author_name":"Wong, Alexander","raw_affiliation_strings":["University of WAterloo"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of WAterloo","institution_ids":["https://openalex.org/I151746483"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5069400644","display_name":"Mahmoud Famouri","orcid":"https://orcid.org/0000-0002-8167-1759","observed_orcids":["https://orcid.org/0000-0002-8167-1759"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Famouri, Mahmoud","raw_affiliation_strings":["DarwinAI Corp. , Waterloo , ON , Canada"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"DarwinAI Corp. , Waterloo , ON , Canada","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5027704717","display_name":"Maya Pavlova","orcid":"https://orcid.org/0000-0003-2348-3650","observed_orcids":["https://orcid.org/0000-0003-2348-3650"]},"institutions":[{"id":"https://openalex.org/I151746483","display_name":"University of Waterloo","ror":"https://ror.org/01aff2v68","country_code":"CA","type":"education","lineage":["https://openalex.org/I151746483"]}],"countries":["CA"],"is_corresponding":false,"raw_author_name":"Pavlova, Maya","raw_affiliation_strings":["DarwinAI Corp. , Waterloo , ON , Canada","Vision and Image Processing Research Group , University of Waterloo , Waterloo , ON , Canada"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"DarwinAI Corp. , Waterloo , ON , Canada","institution_ids":[]},{"raw_affiliation_string":"Vision and Image Processing Research Group , University of Waterloo , Waterloo , ON , Canada","institution_ids":["https://openalex.org/I151746483"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5046744788","display_name":"Siddharth Surana","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Surana, Siddharth","raw_affiliation_strings":["DarwinAI Corp. , Waterloo , ON , Canada"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"DarwinAI Corp. , Waterloo , ON , Canada","institution_ids":[]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":48,"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":false,"primary_topic":{"id":"https://openalex.org/T10201","display_name":"Speech Recognition and Synthesis","score":0.8910999894142151,"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":0.8910999894142151,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.09390000253915787,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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":0.00570000009611249,"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/on-device-speech-recognition","display_name":"on-device speech recognition","score":0.8930000066757202},{"id":"https://openalex.org/keywords/edge-computing","display_name":"edge computing","score":0.8339999914169312},{"id":"https://openalex.org/keywords/tinyml","display_name":"TinyML","score":0.8579999804496765},{"id":"https://openalex.org/keywords/self-attention","display_name":"self-attention","score":0.7770000100135803},{"id":"https://openalex.org/keywords/low-precision-neural-networks","display_name":"low-precision neural networks","score":0.828000009059906},{"id":"https://openalex.org/keywords/model-compression","display_name":"model compression","score":0.7319999933242798},{"id":"https://openalex.org/keywords/microcontroller","display_name":"microcontroller","score":0.6710000038146973},{"id":"https://openalex.org/keywords/embedded-devices","display_name":"embedded devices","score":0.328000009059906}],"concepts":[{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.6376838684082031},{"id":"https://openalex.org/C2984842247","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep neural networks","level":3,"score":0.6214357018470764},{"id":"https://openalex.org/C162307627","wikidata":"https://www.wikidata.org/wiki/Q204833","display_name":"Enhanced Data Rates for GSM Evolution","level":2,"score":0.5916469693183899},{"id":"https://openalex.org/C28490314","wikidata":"https://www.wikidata.org/wiki/Q189436","display_name":"Speech recognition","level":1,"score":0.5873276591300964},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5758675336837769},{"id":"https://openalex.org/C108583219","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep learning","level":2,"score":0.41284289956092834},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.412466824054718}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2008.04245","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2008.04245","pdf_url":"https://arxiv.org/pdf/2008.04245","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2008.04245","doi":"https://doi.org/10.48550/arxiv.2008.04245","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2008.04245","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"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2008.04245","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2008.04245","pdf_url":"https://arxiv.org/pdf/2008.04245","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W3048107166.pdf","grobid_xml":"https://content.openalex.org/works/W3048107166.grobid-xml"},"referenced_works_count":30,"referenced_works":["https://openalex.org/W1922655562","https://openalex.org/W2133564696","https://openalex.org/W2163605009","https://openalex.org/W2407023693","https://openalex.org/W2626778328","https://openalex.org/W2736941579","https://openalex.org/W2752782242","https://openalex.org/W2759071281","https://openalex.org/W2765347007","https://openalex.org/W2766453196","https://openalex.org/W2797583228","https://openalex.org/W2804078698","https://openalex.org/W2808137444","https://openalex.org/W2827311432","https://openalex.org/W2886953980","https://openalex.org/W2888056695","https://openalex.org/W2896457183","https://openalex.org/W2902251695","https://openalex.org/W2919115771","https://openalex.org/W2946928580","https://openalex.org/W2947469743","https://openalex.org/W2949650786","https://openalex.org/W2951559372","https://openalex.org/W2963188574","https://openalex.org/W2964187693","https://openalex.org/W2970783931","https://openalex.org/W2998998803","https://openalex.org/W3132183358","https://openalex.org/W3177096435","https://openalex.org/W7103363508"],"related_works":["https://openalex.org/W2897557658","https://openalex.org/W2975044525","https://openalex.org/W4401042959","https://openalex.org/W7219763761","https://openalex.org/W4382652440","https://openalex.org/W4406058040","https://openalex.org/W4401943448","https://openalex.org/W4372266896","https://openalex.org/W7220094039","https://openalex.org/W3138158912"],"abstract_inverted_index":{"Advances":[0],"in":[1,32],"deep":[2,21,84,128,236],"learning":[3,237],"have":[4],"led":[5],"to":[6,56,207],"state-of-the-art":[7],"performance":[8],"across":[9],"a":[10,29,98,105,142],"multitude":[11],"of":[12,20,77,133,217],"speech":[13,26,57,68,89,139,166,227],"recognition":[14,27,58,90,140,167],"tasks.":[15],"Nevertheless,":[16],"the":[17,36,50,75,92,158,215,239],"widespread":[18],"deployment":[19],"neural":[22,85,129],"networks":[23,86,130,171,224],"for":[24,80,87,137,164,220,225,234],"on-device":[25,88,138,226],"remains":[28],"challenge,":[30],"particularly":[31],"edge":[33,240],"scenarios":[34],"where":[35,49],"memory":[37,51,203],"and":[38,52,103,111,115,193,241],"computing":[39,53],"resources":[40],"are":[41],"highly":[42,222],"constrained":[43],"(e.g.,":[44,61],"low-power":[45],"embedded":[46],"devices)":[47],"or":[48],"budget":[54],"dedicated":[55],"is":[59,97],"low":[60],"mobile":[62],"devices":[63],"performing":[64],"numerous":[65],"tasks":[66],"besides":[67],"recognition).":[69],"In":[70],"this":[71],"study,":[72],"we":[73,124],"introduce":[74,125],"concept":[76],"attention":[78,95,118,134,218],"condensers":[79,135,219],"building":[81,221],"low-footprint,":[82],"highly-efficient":[83],"on":[91,157,238],"edge.":[93],"An":[94],"condenser":[96],"self-attention":[99],"mechanism":[100],"that":[101,169],"learns":[102],"produces":[104],"condensed":[106],"embedding":[107],"characterizing":[108],"joint":[109],"local":[110],"cross-channel":[112],"activation":[113],"relationships,":[114],"performs":[116],"selective":[117],"accordingly.":[119],"To":[120],"illustrate":[121],"its":[122,232],"efficacy,":[123],"TinySpeech,":[126],"low-precision":[127],"comprising":[131],"largely":[132],"tailored":[136,149],"using":[141],"machine-driven":[143],"design":[144],"exploration":[145],"strategy,":[146],"with":[147,151],"one":[148],"specifically":[150],"microcontroller":[152],"operation":[153],"constraints.":[154],"Experimental":[155],"results":[156,211],"Google":[159],"Speech":[160],"Commands":[161],"benchmark":[162],"dataset":[163],"limited-vocabulary":[165],"showed":[168],"TinySpeech":[170],"achieved":[172],"significantly":[173],"lower":[174,183,194,201],"architectural":[175],"complexity":[176,185],"(as":[177,186,197],"much":[178,187,198],"as":[179,188,199],"$507\\times$":[180],"fewer":[181,190],"parameters),":[182],"computational":[184],"$48\\times$":[189],"multiply-add":[191],"operations),":[192],"storage":[195],"requirements":[196],"$2028\\times$":[200],"weight":[202],"requirements)":[204],"when":[205],"compared":[206],"previous":[208],"work.":[209],"These":[210],"not":[212],"only":[213],"demonstrate":[214],"efficacy":[216],"efficient":[223],"recognition,":[228],"but":[229],"also":[230],"illuminate":[231],"potential":[233],"accelerating":[235],"empowering":[242],"TinyML":[243],"applications.":[244]},"counts_by_year":[{"year":2026,"cited_by_count":5},{"year":2025,"cited_by_count":4},{"year":2024,"cited_by_count":8},{"year":2023,"cited_by_count":8},{"year":2022,"cited_by_count":10},{"year":2021,"cited_by_count":9},{"year":2020,"cited_by_count":4}],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2020-08-13T00:00:00"}