{"id":"https://openalex.org/W4408356075","doi":"https://doi.org/10.1109/icassp49660.2025.10889577","title":"Regarding the Existence of the Internal Language Model in CTC-Based E2E ASR","display_name":"Regarding the Existence of the Internal Language Model in CTC-Based E2E ASR","publication_year":2025,"publication_date":"2025-03-12","ids":{"openalex":"https://openalex.org/W4408356075","doi":"https://doi.org/10.1109/icassp49660.2025.10889577"},"language":"en","primary_location":{"id":"doi:10.1109/icassp49660.2025.10889577","is_oa":false,"landing_page_url":"https://doi.org/10.1109/icassp49660.2025.10889577","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":true,"oa_status":"green","oa_url":"https://www.research.ed.ac.uk/files/517113956/ZhaoandBellICASSP2025RegardingtheExistenceoftheInternalLanguageModel.pdf","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5100775936","display_name":"Zeyu Zhao","orcid":"https://orcid.org/0009-0002-6612-9731"},"institutions":[{"id":"https://openalex.org/I98677209","display_name":"University of Edinburgh","ror":"https://ror.org/01nrxwf90","country_code":"GB","type":"education","lineage":["https://openalex.org/I98677209"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Zeyu Zhao","raw_affiliation_strings":["University of Edinburgh,The Centre for Speech Technology Research,Edinburgh,United Kingdom"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Edinburgh,The Centre for Speech Technology Research,Edinburgh,United Kingdom","institution_ids":["https://openalex.org/I98677209"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5081337679","display_name":"Peter Bell","orcid":"https://orcid.org/0000-0003-4415-7408"},"institutions":[{"id":"https://openalex.org/I98677209","display_name":"University of Edinburgh","ror":"https://ror.org/01nrxwf90","country_code":"GB","type":"education","lineage":["https://openalex.org/I98677209"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Peter Bell","raw_affiliation_strings":["University of Edinburgh,The Centre for Speech Technology Research,Edinburgh,United Kingdom"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Edinburgh,The Centre for Speech Technology Research,Edinburgh,United Kingdom","institution_ids":["https://openalex.org/I98677209"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I98677209"],"apc_list":null,"apc_paid":null,"fwci":2.8948,"has_fulltext":true,"cited_by_count":2,"citation_normalized_percentile":{"value":0.90277167,"is_in_top_1_percent":false,"is_in_top_10_percent":true},"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/T10181","display_name":"Natural Language Processing Techniques","score":0.9521999955177307,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.9521999955177307,"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":0.7457674145698547},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.4426558017730713},{"id":"https://openalex.org/keywords/natural-language-processing","display_name":"Natural language processing","score":0.3136652708053589}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7457674145698547},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.4426558017730713},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.3136652708053589}],"mesh":[],"locations_count":2,"locations":[{"id":"doi:10.1109/icassp49660.2025.10889577","is_oa":false,"landing_page_url":"https://doi.org/10.1109/icassp49660.2025.10889577","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"},{"id":"pmh:oai:pure.ed.ac.uk:publications/bc45571b-1734-453f-adb8-bc387f41e0b9","is_oa":true,"landing_page_url":"https://www.research.ed.ac.uk/en/publications/bc45571b-1734-453f-adb8-bc387f41e0b9","pdf_url":"https://www.research.ed.ac.uk/files/517113956/ZhaoandBellICASSP2025RegardingtheExistenceoftheInternalLanguageModel.pdf","source":{"id":"https://openalex.org/S4406922455","display_name":"Edinburgh Research Explorer","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"Zhao, Z & Bell, P 2025, Regarding the existence of the internal language model in CTC-based E2E ASR. in Proceedings of the 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). Proceedings of the IEEE International Conference on Acoustics, Speech, and Signal Processing, Institute of Electrical and Electronics Engineers, Piscataway, NJ, USA, pp. 1-5, 2025 IEEE International Conference on Acoustics, Speech, and Signal Processing, Hyderabad, India, 6/04/25. https://doi.org/10.1109/ICASSP49660.2025.10889577","raw_type":"info:eu-repo/semantics/conferenceObject"}],"best_oa_location":{"id":"pmh:oai:pure.ed.ac.uk:publications/bc45571b-1734-453f-adb8-bc387f41e0b9","is_oa":true,"landing_page_url":"https://www.research.ed.ac.uk/en/publications/bc45571b-1734-453f-adb8-bc387f41e0b9","pdf_url":"https://www.research.ed.ac.uk/files/517113956/ZhaoandBellICASSP2025RegardingtheExistenceoftheInternalLanguageModel.pdf","source":{"id":"https://openalex.org/S4406922455","display_name":"Edinburgh Research Explorer","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"Zhao, Z & Bell, P 2025, Regarding the existence of the internal language model in CTC-based E2E ASR. in Proceedings of the 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). Proceedings of the IEEE International Conference on Acoustics, Speech, and Signal Processing, Institute of Electrical and Electronics Engineers, Piscataway, NJ, USA, pp. 1-5, 2025 IEEE International Conference on Acoustics, Speech, and Signal Processing, Hyderabad, India, 6/04/25. https://doi.org/10.1109/ICASSP49660.2025.10889577","raw_type":"info:eu-repo/semantics/conferenceObject"},"sustainable_development_goals":[{"display_name":"Quality Education","score":0.47999998927116394,"id":"https://metadata.un.org/sdg/4"}],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4408356075.pdf","grobid_xml":"https://content.openalex.org/works/W4408356075.grobid-xml"},"referenced_works_count":27,"referenced_works":["https://openalex.org/W2005708641","https://openalex.org/W2127141656","https://openalex.org/W2144499799","https://openalex.org/W2327501763","https://openalex.org/W2402146185","https://openalex.org/W2766219058","https://openalex.org/W2808939837","https://openalex.org/W2888867175","https://openalex.org/W2962780374","https://openalex.org/W2963211739","https://openalex.org/W2963250244","https://openalex.org/W2963537349","https://openalex.org/W3016053754","https://openalex.org/W3152221657","https://openalex.org/W3163560333","https://openalex.org/W3205788551","https://openalex.org/W3209059054","https://openalex.org/W4319862418","https://openalex.org/W4372266720","https://openalex.org/W4372267353","https://openalex.org/W4385823095","https://openalex.org/W4385823471","https://openalex.org/W4391021728","https://openalex.org/W6631362777","https://openalex.org/W6679855610","https://openalex.org/W6780218876","https://openalex.org/W6847363464"],"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":{"Some":[0],"End-to-End":[1],"(E2E)":[2],"Automatic":[3],"Speech":[4],"Recognition":[5],"(ASR)":[6],"models,":[7,103],"such":[8],"as":[9,27],"Attention-based":[10],"Encoder-Decoder":[11],"(AED)":[12],"and":[13,64,122],"Recurrent":[14],"Neural":[15],"Network":[16],"Transducer":[17],"(RNN-T)":[18],"are":[19],"known":[20],"to":[21,86,107],"have":[22],"components":[23],"that":[24,155],"effectively":[25],"act":[26],"internal":[28],"language":[29,102],"models":[30,109],"(ILM),":[31],"implicitly":[32],"modelling":[33,120],"the":[34,38,42,62,88,115,142,152,173],"prior":[35],"probability":[36],"of":[37,44,66,117,133],"output":[39],"sequence.":[40],"However,":[41,149],"existence":[43,63],"an":[45,67,134,164,167],"ILM":[46,68,135,168],"in":[47,69,136,146,178],"pure":[48],"Connectionist":[49],"Temporal":[50],"Classification":[51],"(CTC)":[52],"ASR":[53,138],"systems":[54],"remains":[55],"debated.":[56],"In":[57],"this":[58],"paper,":[59],"we":[60,80,104,150],"investigate":[61],"strength":[65],"CTC":[70,73,108,158,174],"systems.":[71],"Since":[72],"posterior":[74],"probabilities":[75],"cannot":[76],"be":[77],"analytically":[78],"factorised,":[79],"propose":[81],"a":[82,95,157],"novel":[83],"empirical":[84],"method":[85,93],"probe":[87],"ILM.":[89],"After":[90],"validating":[91],"our":[92,147],"on":[94],"hybrid":[96],"DNN":[97],"model":[98],"with":[99,141,163],"various":[100],"external":[101],"apply":[105],"it":[106],"trained":[110,162],"under":[111],"different":[112],"conditions,":[113],"examining":[114],"effects":[116],"training":[118,123,144],"data,":[119],"units,":[121],"or":[124],"pre-training":[125],"methods.":[126],"Our":[127],"results":[128],"show":[129],"no":[130],"strong":[131],"evidence":[132],"CTC-based":[137],"systems,":[139],"even":[140,170],"largest":[143],"dataset":[145],"experiments.":[148],"make":[151],"surprising":[153],"finding":[154],"when":[156,171],"encoder":[159],"is":[160,176],"jointly":[161],"AED":[165],"loss,":[166],"emerges,":[169],"only":[172],"component":[175],"used":[177],"decoding.":[179]},"counts_by_year":[{"year":2025,"cited_by_count":2}],"updated_date":"2026-07-03T08:13:44.112507","created_date":"2025-10-10T00:00:00"}