{"id":"https://openalex.org/W194138933","doi":"https://doi.org/10.21437/interspeech.2010-128","title":"Accelerating hierarchical acoustic likelihood computation on graphics processors","display_name":"Accelerating hierarchical acoustic likelihood computation on graphics processors","publication_year":2010,"publication_date":"2010-09-26","ids":{"openalex":"https://openalex.org/W194138933","doi":"https://doi.org/10.21437/interspeech.2010-128","mag":"194138933"},"language":"en","primary_location":{"id":"doi:10.21437/interspeech.2010-128","is_oa":false,"landing_page_url":"https://doi.org/10.21437/interspeech.2010-128","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Interspeech 2010","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/A5051360374","display_name":"Pavel Kv\u011bto\u0148","orcid":null},"institutions":[{"id":"https://openalex.org/I1341412227","display_name":"IBM (United States)","ror":"https://ror.org/05hh8d621","country_code":"US","type":"company","lineage":["https://openalex.org/I1341412227"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Pavel Kv\u011bto\u0148","raw_affiliation_strings":["IBM, ,"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"IBM, ,","institution_ids":["https://openalex.org/I1341412227"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5030825867","display_name":"Miroslav Nov\u00e1k","orcid":"https://orcid.org/0000-0001-9239-4120"},"institutions":[{"id":"https://openalex.org/I1341412227","display_name":"IBM (United States)","ror":"https://ror.org/05hh8d621","country_code":"US","type":"company","lineage":["https://openalex.org/I1341412227"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Miroslav Nov\u00e1k","raw_affiliation_strings":["IBM, ,"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"IBM, ,","institution_ids":["https://openalex.org/I1341412227"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I1341412227"],"apc_list":null,"apc_paid":null,"fwci":1.6348,"has_fulltext":false,"cited_by_count":8,"citation_normalized_percentile":{"value":0.83276079,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":89,"max":98},"biblio":{"volume":null,"issue":null,"first_page":"350","last_page":"353"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10201","display_name":"Speech Recognition and Synthesis","score":0.9997000098228455,"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":0.9997000098228455,"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/T10901","display_name":"Advanced Data Compression Techniques","score":0.9994000196456909,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.8210420608520508},{"id":"https://openalex.org/keywords/computation","display_name":"Computation","score":0.7285738587379456},{"id":"https://openalex.org/keywords/parallel-computing","display_name":"Parallel computing","score":0.6058393716812134},{"id":"https://openalex.org/keywords/graphics-processing-unit","display_name":"Graphics processing unit","score":0.5729687213897705},{"id":"https://openalex.org/keywords/gaussian","display_name":"Gaussian","score":0.544553279876709},{"id":"https://openalex.org/keywords/pruning","display_name":"Pruning","score":0.5206679105758667},{"id":"https://openalex.org/keywords/matrix-multiplication","display_name":"Matrix multiplication","score":0.4823529124259949},{"id":"https://openalex.org/keywords/graphics","display_name":"Graphics","score":0.46459436416625977},{"id":"https://openalex.org/keywords/multiplication","display_name":"Multiplication (music)","score":0.46107912063598633},{"id":"https://openalex.org/keywords/central-processing-unit","display_name":"Central processing unit","score":0.4607826769351959},{"id":"https://openalex.org/keywords/gaussian-elimination","display_name":"Gaussian elimination","score":0.4537619948387146},{"id":"https://openalex.org/keywords/cuda","display_name":"CUDA","score":0.4161488711833954},{"id":"https://openalex.org/keywords/algorithm","display_name":"Algorithm","score":0.38167139887809753},{"id":"https://openalex.org/keywords/computer-hardware","display_name":"Computer hardware","score":0.15884879231452942},{"id":"https://openalex.org/keywords/mathematics","display_name":"Mathematics","score":0.097665935754776},{"id":"https://openalex.org/keywords/computer-graphics","display_name":"Computer graphics (images)","score":0.07051819562911987}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8210420608520508},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.7285738587379456},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.6058393716812134},{"id":"https://openalex.org/C2779851693","wikidata":"https://www.wikidata.org/wiki/Q183484","display_name":"Graphics processing unit","level":2,"score":0.5729687213897705},{"id":"https://openalex.org/C163716315","wikidata":"https://www.wikidata.org/wiki/Q901177","display_name":"Gaussian","level":2,"score":0.544553279876709},{"id":"https://openalex.org/C108010975","wikidata":"https://www.wikidata.org/wiki/Q500094","display_name":"Pruning","level":2,"score":0.5206679105758667},{"id":"https://openalex.org/C17349429","wikidata":"https://www.wikidata.org/wiki/Q1049914","display_name":"Matrix multiplication","level":3,"score":0.4823529124259949},{"id":"https://openalex.org/C21442007","wikidata":"https://www.wikidata.org/wiki/Q1027879","display_name":"Graphics","level":2,"score":0.46459436416625977},{"id":"https://openalex.org/C2780595030","wikidata":"https://www.wikidata.org/wiki/Q3860309","display_name":"Multiplication (music)","level":2,"score":0.46107912063598633},{"id":"https://openalex.org/C49154492","wikidata":"https://www.wikidata.org/wiki/Q5300","display_name":"Central processing unit","level":2,"score":0.4607826769351959},{"id":"https://openalex.org/C126312332","wikidata":"https://www.wikidata.org/wiki/Q2658","display_name":"Gaussian elimination","level":3,"score":0.4537619948387146},{"id":"https://openalex.org/C2778119891","wikidata":"https://www.wikidata.org/wiki/Q477690","display_name":"CUDA","level":2,"score":0.4161488711833954},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.38167139887809753},{"id":"https://openalex.org/C9390403","wikidata":"https://www.wikidata.org/wiki/Q3966","display_name":"Computer hardware","level":1,"score":0.15884879231452942},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.097665935754776},{"id":"https://openalex.org/C121684516","wikidata":"https://www.wikidata.org/wiki/Q7600677","display_name":"Computer graphics (images)","level":1,"score":0.07051819562911987},{"id":"https://openalex.org/C86803240","wikidata":"https://www.wikidata.org/wiki/Q420","display_name":"Biology","level":0,"score":0.0},{"id":"https://openalex.org/C121332964","wikidata":"https://www.wikidata.org/wiki/Q413","display_name":"Physics","level":0,"score":0.0},{"id":"https://openalex.org/C62520636","wikidata":"https://www.wikidata.org/wiki/Q944","display_name":"Quantum mechanics","level":1,"score":0.0},{"id":"https://openalex.org/C6557445","wikidata":"https://www.wikidata.org/wiki/Q173113","display_name":"Agronomy","level":1,"score":0.0},{"id":"https://openalex.org/C84114770","wikidata":"https://www.wikidata.org/wiki/Q46344","display_name":"Quantum","level":2,"score":0.0},{"id":"https://openalex.org/C114614502","wikidata":"https://www.wikidata.org/wiki/Q76592","display_name":"Combinatorics","level":1,"score":0.0}],"mesh":[],"locations_count":2,"locations":[{"id":"doi:10.21437/interspeech.2010-128","is_oa":false,"landing_page_url":"https://doi.org/10.21437/interspeech.2010-128","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Interspeech 2010","raw_type":"proceedings-article"},{"id":"mag:194138933","is_oa":false,"landing_page_url":"https://dblp.uni-trier.de/db/conf/interspeech/interspeech2010.html#KvetonN10","pdf_url":null,"source":null,"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":"Conference of the International Speech Communication Association","raw_type":null}],"best_oa_location":null,"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":8,"referenced_works":["https://openalex.org/W60145987","https://openalex.org/W167388389","https://openalex.org/W2037102844","https://openalex.org/W2097003869","https://openalex.org/W2127095586","https://openalex.org/W2160688560","https://openalex.org/W2333081718","https://openalex.org/W2489815359"],"related_works":["https://openalex.org/W2037102844","https://openalex.org/W34089377","https://openalex.org/W1963866266","https://openalex.org/W2339588615","https://openalex.org/W2945926249","https://openalex.org/W2751155736","https://openalex.org/W2253442234","https://openalex.org/W2549366874","https://openalex.org/W2313264472","https://openalex.org/W3006491016","https://openalex.org/W2959982948","https://openalex.org/W2239144794","https://openalex.org/W3003818908","https://openalex.org/W2893346234","https://openalex.org/W3128210089","https://openalex.org/W3111098492","https://openalex.org/W2884212285","https://openalex.org/W2902494083","https://openalex.org/W2918317383","https://openalex.org/W2065392434"],"abstract_inverted_index":{"pavel.kveton@cz.ibm.com,":[0],"miroslav@us.ibm.comABSTRACTThe":[1],"paper":[2,45,117,449,519,552],"presents":[3,346,450],"a":[4,15,25,35,61,89,127,145,158,183,199,225,237,334,347,362,399,431,564,571,604,748,775],"method":[5,65],"for":[6,40,76,161,233,264,272,294,310,438],"performance":[7,100,172,690,750],"improvements":[8],"ofa":[9],"speech":[10,315,365],"recognition":[11,143,366],"system":[12],"by":[13,101,210,419,619,771],"moving":[14],"part":[16],"of":[17,48,51,57,80,153,165,173,178,213,242,280,299,321,390,436,453,481,491,576,622,646,670,680],"the":[18,30,103,116,154,171,179,211,297,302,391,410,475,518,545,555,580,620,624,629,647,681,688,720,739,744,766,772],"computation":[19,22,56,71,87,104,509],"-acoustic":[20],"likelihood":[21,53],"-":[23,560,562],"onto":[24],"Graphics":[26],"Processor":[27],"Unit(GPU).":[28],"In":[29],"system,":[31],"GPU":[32,77,128,246,335,348,377,411,432,457,467,572],"operates":[33],"as":[34,653,702,704],"low":[36],"cost":[37],"powerful":[38,308],"co-processor":[39],"linear":[41,311],"algebra":[42],"operations.":[43],"The":[44,85,583,587,752],"compares":[46,459],"GPUimplementation":[47],"two":[49],"techniques":[50,107],"acoustic":[52,174],"computa-tion:":[54],"fullGaussian":[55],"all":[58,623,632,711,729],"components":[59],"and":[60,283,287,292,307,458,468,602,606,639,657,758],"signi\ufb01cantlyfaster":[62],"Gaussian":[63,70,135,185,326,351,374,402,464,476,508],"selection":[64,477],"using":[66,126],"hierarchical":[67,120,184,454],"evaluation.":[68],"Thefull":[69],"is":[72,88,144,182,329,354,483,550,563,568,585,612,660,664,685,695,727,760],"an":[73,268,305,451],"ideal":[74,761],"candidate":[75,160],"implemen-tation":[78],"because":[79,548],"its":[81,338],"matrix":[82,339],"multiplication":[83,340],"nature.":[84],"hierarchicalGaussian":[86],"technique":[90],"commonly":[91],"used":[92,522],"on":[93,176,333,361,376,446,456,466,487,510,529,570,593,691,719],"aCPU":[94],"sinceit":[95],"leads":[96,386,499],"to":[97,112,169,189,260,324,331,337,387,500,628,762],"much":[98,110,130,703],"better":[99,131],"pruning":[102,406,480],"vol-ume.":[105],"Pruning":[106],"are":[108,673],"generally":[109],"harder":[111],"implementon":[113],"GPUs,":[114],"nevertheless,":[115],"shows":[118],"that":[119,229,353,498,567,672],"Gaussiancomputation":[121],"can":[122,206,599,650],"still":[123],"be":[124,207,600,651],"done":[125],"with":[129,195,405,424,461,469,506,525,747],"perfor-mance":[132,204],"than":[133,357,415],"full":[134,196,325,350,373,463,507],"computation.Index":[136],"Terms\u2014":[137],"Speech":[138],"recognition,":[139],"GPU,":[140,488],"Parallelization1.":[141],"INTRODUCTIONSpeech":[142],"CPU-intensive":[146],"tasks.":[147,266],"Acoustic":[148],"likelihoodevaluation":[149],"usually":[150],"represents":[151],"30-50%":[152],"computation,":[155],"whichmakes":[156],"it":[157,258,460,549,659,694,759],"good":[159],"optimization.":[162],"A":[163,643],"lot":[164],"work":[166],"has":[167,230,275,495],"beendedicated":[168],"increase":[170,205],"scoring":[175],"CPU.One":[177],"successful":[180],"methods":[181],"selection[1,":[186],"2]":[187],"leading":[188],"3-4x":[190],"speed":[191,503],"up":[192,389,504],"in":[193,278,314,433,573,589,706,731,735,738],"comparison":[194],"Gaussianevaluation":[197,455],"without":[198],"noteworthy":[200],"accuracy":[201],"loss.":[202],"Signi\ufb01cant":[203],"also":[208,263,313],"achieved":[209],"use":[212,261,697],"Streaming":[214],"SIMDExtension":[215],"(SSE)":[216],"instruction":[217,741],"set":[218],"[2,":[219],"3].Graphical":[220],"processor":[221],"units":[222],"(GPUs)":[223],"represent":[224],"low-cost":[226],"computa-tional":[227,535],"power":[228],"been":[231,276,496],"reserved":[232],"image/video":[234],"processing":[235],"tasksfor":[236],"long":[238],"time.":[239],"An":[240],"introduction":[241],"General":[243],"Purpose":[244],"computationson":[245],"(GPGPU,":[247],"e.g.":[248],"CUDA":[249,526],"from":[250],"NVIDIA":[251,286,523],"[4],":[252],"ATI":[253],"Stream":[254],"fromAMD":[255],"[5])":[256],"made":[257],"feasible":[259],"GPUs":[262,322,524],"other":[265],"Re-cently,":[267],"OpenCL":[269,543,556,558],"[6]":[270],"standard":[271],"cross-platform":[273],"parallelprogram-ming":[274],"developed":[277],"cooperation":[279],"many":[281],"industry-leadingcompanies":[282],"institutions.":[284],"Both":[285],"AMD":[288],"provide":[289],"anOpenCL":[290],"SDK":[291],"drivers":[293],"their":[295,715],"GPUs.With":[296],"dawn":[298],"GPGPU":[300],"environments":[301],"graphics":[303],"processorsbecome":[304],"affordable":[306],"coprocessors":[309],"algebraoperations":[312],"recognition.":[316],"Recent":[317],"works":[318],"show":[319],"contribu-tion":[320],"mainly":[323],"evaluation,":[327,384],"which":[328,385],"straight-forward":[330],"implement":[332],"due":[336],"na-ture.":[341],"Dixon":[342],"at":[343],"al.":[344,370,396,422,429],"[7]":[345],"based":[349,528],"evalua-tion":[352,465],"4\u20136x":[355],"faster":[356,379,414],"on-demand":[358,383,401],"CPU":[359,382,472,493],"evaluation":[360,375,403],"largevocabulary":[363],"continuous":[364],"task":[367],"(LVCSR).":[368],"Cardinalat":[369],"[3]":[371],"reports":[372],"5x":[378],"thanan":[380],"SSE-optimized":[381],"35%speed":[388],"LVCSR":[392],"task.":[393],"Chong":[394],"et":[395,421,428],"[8]":[397],"implemented":[398],"model-optimized":[400],"together":[404],"theViterbi":[407],"search":[408],"showing":[409],"approx.":[412,501],"9x":[413],"CPU,":[416],"gener-alized":[417],"later":[418],"You":[420],"[9]":[423],"comparable":[425],"results.":[426],"Shi":[427],"[10]employs":[430],"\ufb01nding":[434],"clusters":[435],"Gaussians":[437],"fMPE":[439],"discri":[440],"mi-native":[441],"training":[442],"achieving":[443],"17x":[444],"speed-up":[445],"this":[447,725],"task.This":[448],"implementation":[452],"both":[462],"SSE":[470],"optimized":[471],"implementations.":[473],"Al-though":[474],"requires":[478,768],"global":[479,648],"clusterswhich":[482],"not":[484],"very":[485],"ef\ufb01cient":[486],"acceptable":[489],"combination":[490],"GPUand":[492],"usage":[494],"found":[497],"2x":[502],"incomparison":[505],"GPU.2.":[511],"SOFTAW":[512],"RE":[513,516],"AND":[514],"HARDAW":[515],"ENVIRONMENTIn":[517],"we":[520,540],"have":[521,541,634,714],"architecture[4]":[527],"SIMT":[530],"(Single":[531],"Instruction":[532],"Multiple":[533],"Thread)":[534],"model.":[536],"Among":[537],"several":[538],"options,":[539],"chosen":[542],"[6]as":[544],"programming":[546],"framework":[547],"platform-independent.The":[551],"hence":[553],"uses":[554],"terminology.An":[557],"program":[559],"akernel":[561],"code":[565],"segment":[566],"exe-cuted":[569],"N-Dimensional":[574],"range(NDrange)":[575],"work-items(all":[577],"work-items":[578,588,633,671,682,730,773],"execute":[579],"same":[581],"code).":[582],"NDrange":[584],"splitintoequally-sizedwork-groups.":[586],"one":[590,594],"work-group":[591,663],"runtogether":[592],"streaming":[595],"multiprocessor":[596],"where":[597],"they":[598],"syn-chronized":[601],"share":[603],"small":[605,644],"low-latencylocal":[607],"memory":[608,637,649,678,701,712,734,754],"(typi-cally":[609],"16kB).":[610],"There":[611],"no":[613],"communication":[614],"available":[615],"between":[616],"differentwork-groups":[617],"except":[618],"termination":[621],"work-itemsin":[625],"theNDrange.In":[626],"addition":[627],"local":[630,698,753],"memory,":[631,709],"access":[635,679,733,745,770],"toglobal":[636],"(large":[638],"high-latency,":[640],"typically":[641],"1GB).":[642],"por-tion":[645],"declared":[652],"constant":[654,700],"(read-only,typically":[655],"64kB)":[656],"then":[658],"cashed":[661],".Each":[662],"split":[665],"intohalf-warps(two":[666],"half-warps":[667,684],"forma":[668],"warp)":[669],"executed":[674],"together.":[675],"Optimization":[676],"ofthe":[677],"within":[683,774],"crucial":[686],"toachieve":[687],"best":[689,740],"GPU.":[692],"Basically,":[693],"recommendedto":[696],"or":[699],"possible":[705],"favor":[707],"ofglobal":[708],"however,":[710],"spaces":[713],"optimal":[716],"ac-cess":[717],"strategy":[718,726],"CUDA-enabled":[721],"chips":[722],"\u2014":[723],"if":[724],"fol-lowed,":[728],"half-warp":[732],"parallel":[736],"re-sulting":[737],"throughput.":[742],"Otherwise,":[743],"isserialized":[746],"serious":[749],"impact.":[751],"isorganized":[755],"into":[756],"banks":[757],"avoid":[763],"bank":[764],"con\ufb02icts;":[765],"globalmemory":[767],"coalescent":[769],"half-":[776]},"counts_by_year":[{"year":2016,"cited_by_count":1},{"year":2014,"cited_by_count":1},{"year":2012,"cited_by_count":4}],"updated_date":"2026-07-29T14:22:42.915294","created_date":"2025-10-10T00:00:00"}