{"id":"https://openalex.org/W7154734763","doi":"https://doi.org/10.48550/arxiv.2604.14493","title":"Pushing the Limits of On-Device Streaming ASR: A Compact, High-Accuracy English Model for Low-Latency Inference","display_name":"Pushing the Limits of On-Device Streaming ASR: A Compact, High-Accuracy English Model for Low-Latency Inference","publication_year":2026,"publication_date":"2026-04-16","ids":{"openalex":"https://openalex.org/W7154734763","doi":"https://doi.org/10.48550/arxiv.2604.14493"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2604.14493","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.14493","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2604.14493","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5065383769","display_name":"Nenad Banfi\u0107","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Banfic, Nenad","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133863067","display_name":"David Fan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Fan, David","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133845517","display_name":"Kunal Vaishnavi","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Vaishnavi, Kunal","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133879386","display_name":"Sam Kemp","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kemp, Sam","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133857333","display_name":"Sunghoon Choi","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Choi, Sunghoon","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133891674","display_name":"Rui Ren","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ren, Rui","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133832268","display_name":"Sayan Shaw","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Shaw, Sayan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5133828073","display_name":"Meng Tang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tang, Meng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"cited_by_count":0,"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":0.8138999938964844,"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":0.8138999938964844,"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":0.06390000134706497,"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":0.03889999911189079,"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/streaming-algorithm","display_name":"Streaming algorithm","score":0.6812999844551086},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.5975000262260437},{"id":"https://openalex.org/keywords/benchmark","display_name":"Benchmark (surveying)","score":0.5404000282287598},{"id":"https://openalex.org/keywords/pipeline","display_name":"Pipeline (software)","score":0.4781000018119812},{"id":"https://openalex.org/keywords/memory-footprint","display_name":"Memory footprint","score":0.439300000667572},{"id":"https://openalex.org/keywords/quantization","display_name":"Quantization (signal processing)","score":0.4163999855518341},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.4081000089645386},{"id":"https://openalex.org/keywords/point","display_name":"Point (geometry)","score":0.40230000019073486}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8122000098228455},{"id":"https://openalex.org/C187166803","wikidata":"https://www.wikidata.org/wiki/Q2835831","display_name":"Streaming algorithm","level":3,"score":0.6812999844551086},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5975000262260437},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.5404000282287598},{"id":"https://openalex.org/C43521106","wikidata":"https://www.wikidata.org/wiki/Q2165493","display_name":"Pipeline (software)","level":2,"score":0.4781000018119812},{"id":"https://openalex.org/C74912251","wikidata":"https://www.wikidata.org/wiki/Q6815727","display_name":"Memory footprint","level":2,"score":0.439300000667572},{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.4163999855518341},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.4081000089645386},{"id":"https://openalex.org/C28719098","wikidata":"https://www.wikidata.org/wiki/Q44946","display_name":"Point (geometry)","level":2,"score":0.40230000019073486},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.35659998655319214},{"id":"https://openalex.org/C84211073","wikidata":"https://www.wikidata.org/wiki/Q117879","display_name":"Floating point","level":2,"score":0.3553999960422516},{"id":"https://openalex.org/C138236772","wikidata":"https://www.wikidata.org/wiki/Q25098575","display_name":"Edge device","level":3,"score":0.3472000062465668},{"id":"https://openalex.org/C90805587","wikidata":"https://www.wikidata.org/wiki/Q10944557","display_name":"Word (group theory)","level":2,"score":0.34610000252723694},{"id":"https://openalex.org/C162307627","wikidata":"https://www.wikidata.org/wiki/Q204833","display_name":"Enhanced Data Rates for GSM Evolution","level":2,"score":0.33309999108314514},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.32820001244544983},{"id":"https://openalex.org/C28490314","wikidata":"https://www.wikidata.org/wiki/Q189436","display_name":"Speech recognition","level":1,"score":0.3163999915122986},{"id":"https://openalex.org/C149810388","wikidata":"https://www.wikidata.org/wiki/Q5374873","display_name":"Emulation","level":2,"score":0.3158000111579895},{"id":"https://openalex.org/C134535813","wikidata":"https://www.wikidata.org/wiki/Q1888734","display_name":"Transcoding","level":2,"score":0.30709999799728394},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.30059999227523804},{"id":"https://openalex.org/C160234255","wikidata":"https://www.wikidata.org/wiki/Q812535","display_name":"Bayesian inference","level":3,"score":0.2919999957084656},{"id":"https://openalex.org/C40969351","wikidata":"https://www.wikidata.org/wiki/Q3516228","display_name":"Word error rate","level":2,"score":0.27900001406669617},{"id":"https://openalex.org/C137635306","wikidata":"https://www.wikidata.org/wiki/Q182667","display_name":"Pareto principle","level":2,"score":0.2736999988555908},{"id":"https://openalex.org/C94966114","wikidata":"https://www.wikidata.org/wiki/Q29256","display_name":"Black box","level":2,"score":0.2603999972343445}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2604.14493","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.14493","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2604.14493","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.14493","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Deploying":[0],"high-quality":[1],"automatic":[2],"speech":[3],"recognition":[4],"(ASR)":[5],"on":[6,22,85,171],"edge":[7],"devices":[8],"requires":[9],"models":[10],"that":[11],"jointly":[12],"optimize":[13],"accuracy,":[14],"latency,":[15,177],"and":[16,40,47,69,99,114],"memory":[17],"footprint":[18],"while":[19,136],"operating":[20],"entirely":[21],"CPU":[23,172],"without":[24],"GPU":[25],"acceleration.":[26],"We":[27,88],"conduct":[28,100],"a":[29,52,101,179],"systematic":[30],"empirical":[31],"study":[32],"of":[33,55,104,145],"state-of-the-art":[34],"ASR":[35],"architectures,":[36],"encompassing":[37],"encoder-decoder,":[38],"transducer,":[39],"LLM-based":[41],"paradigms,":[42],"evaluated":[43],"across":[44,162],"batch,":[45],"chunked,":[46],"streaming":[48,84,93,160,186],"inference":[49,94],"modes.":[50],"Through":[51],"comprehensive":[53],"benchmark":[54],"over":[56],"50":[57],"configurations":[58],"spanning":[59],"OpenAI":[60],"Whisper,":[61],"NVIDIA":[62],"Nemotron,":[63],"Parakeet":[64],"TDT,":[65],"Canary,":[66],"Conformer":[67],"Transducer,":[68],"Qwen3-ASR,":[70],"we":[71],"identify":[72],"NVIDIA's":[73],"Nemotron":[74],"Speech":[75],"Streaming":[76],"as":[77,131,133],"the":[78,91,125,146,153],"strongest":[79],"candidate":[80],"for":[81,184],"real-time":[82,170],"English":[83],"resource-constrained":[86],"hardware.":[87],"then":[89],"re-implement":[90],"complete":[92],"pipeline":[95],"in":[96],"ONNX":[97],"Runtime":[98],"controlled":[102],"evaluation":[103],"multiple":[105],"post-training":[106],"quantization":[107],"strategies,":[108],"including":[109],"importance-weighted":[110],"k-quant,":[111],"mixed-precision":[112],"schemes,":[113],"round-to-nearest":[115],"quantization,":[116],"combined":[117],"with":[118,173],"graph-level":[119],"operator":[120],"fusion.":[121],"These":[122],"optimizations":[123],"reduce":[124],"model":[126],"from":[127],"2.47":[128],"GB":[129,135],"to":[130],"little":[132],"0.67":[134],"maintaining":[137],"word":[138],"error":[139],"rate":[140],"(WER)":[141],"within":[142],"1%":[143],"absolute":[144],"full-precision":[147],"PyTorch":[148],"baseline.":[149],"Our":[150],"recommended":[151],"configuration,":[152],"int4":[154],"k-quant":[155],"variant,":[156],"achieves":[157],"8.20%":[158],"average":[159],"WER":[161],"eight":[163],"standard":[164],"benchmarks,":[165],"running":[166],"comfortably":[167],"faster":[168],"than":[169],"0.56":[174],"s":[175],"algorithmic":[176],"establishing":[178],"new":[180],"quality-efficiency":[181],"Pareto":[182],"point":[183],"on-device":[185],"ASR.":[187]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-04-18T00:00:00"}