{"corpus_id":49862364,"paper_sha":"026cbd39b90daa893d1c080dc43461f9f4f9d977","doi":null,"arxiv_id":"1807.05855","pmid":null,"pmcid":null,"mag_id":2883762569,"dblp_id":"journals/corr/abs-1807-05855","acl_id":null,"title":"A Fast-Converged Acoustic Modeling for Korean Speech Recognition: A Preliminary Study on Time Delay Neural Network","year":2018,"publication_date":"2018-07-11","venue":"arXiv.org","journal":{"name":"ArXiv","pages":null,"volume":"abs/1807.05855"},"journal_issn":null,"journal_title":null,"publication_types":["JournalArticle"],"pubmed_pub_types":null,"s2_fields_of_study":["Computer Science","Engineering"],"reference_count":13,"citation_count":2,"influential_citation_count":0,"is_open_access":false,"arxiv_categories":["cs.CL","cs.SD","eess.AS"],"arxiv_license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","arxiv_journal_ref":null,"mesh_headings":null,"chemicals":null,"comments_corrections":null,"source_flags":1,"s2_open_access_pdf_url":null,"s2_open_access_landing_url":null,"s2_open_access_license":null,"s2_open_access_status":null,"pmc_open_access_pdf_url":null,"pmc_open_access_landing_url":null,"pmc_open_access_license":null,"pmc_open_access_status":null,"unpaywall_open_access_pdf_url":null,"unpaywall_open_access_landing_url":null,"unpaywall_open_access_license":null,"unpaywall_open_access_status":null,"abstract":"In this paper, a time delay neural network (TDNN) based acoustic model is proposed to implement a fast-converged acoustic modeling for Korean speech recognition. The TDNN has an advantage in fast-convergence where the amount of training data is limited, due to subsampling which excludes duplicated weights. The TDNN showed an absolute improvement of 2.12% in terms of character error rate compared to feed forward neural network (FFNN) based modelling for Korean speech corpora. The proposed model converged 1.67 times faster than a FFNN-based model did.","claims":[{"public_id":"cl_fefb2bbef48702c72db9e62c06c1c67f","status":"active","text":"The TDNN has an advantage in fast convergence when training data is limited, due to subsampling which excludes duplicated weights.","confidence":0.9,"contributors":[{"id":2,"public_id":"4715169a40","public_label":"AK (4715169a40)","roles":["extraction"],"url":"https://sah.borca.ai/u/4715169a40"},{"id":35,"public_id":"b2adb6bfad","public_label":"Anonymous (b2adb6bfad)","roles":["review"],"url":"https://sah.borca.ai/u/b2adb6bfad"},{"id":136,"public_id":"3c2apqe3ut","public_label":"Anonymous (3c2apqe3ut)","roles":["review"],"url":"https://sah.borca.ai/u/3c2apqe3ut"}],"url":"https://sah.borca.ai/claims/cl_fefb2bbef48702c72db9e62c06c1c67f"},{"public_id":"cl_0cfde1214911a2ff4e0baee70400eadb","status":"active","text":"The proposed TDNN-based acoustic model achieves an absolute improvement of 2.12% in character error rate compared to FFNN-based modeling for Korean speech corpora.","confidence":0.95,"contributors":[{"id":2,"public_id":"4715169a40","public_label":"AK (4715169a40)","roles":["extraction"],"url":"https://sah.borca.ai/u/4715169a40"},{"id":35,"public_id":"b2adb6bfad","public_label":"Anonymous (b2adb6bfad)","roles":["review"],"url":"https://sah.borca.ai/u/b2adb6bfad"},{"id":136,"public_id":"3c2apqe3ut","public_label":"Anonymous (3c2apqe3ut)","roles":["review"],"url":"https://sah.borca.ai/u/3c2apqe3ut"}],"url":"https://sah.borca.ai/claims/cl_0cfde1214911a2ff4e0baee70400eadb"},{"public_id":"cl_33d25f5bda751b8f891625fd16ee10d8","status":"active","text":"The proposed TDNN-based acoustic model converges 1.67 times faster than a FFNN-based model.","confidence":0.95,"contributors":[{"id":2,"public_id":"4715169a40","public_label":"AK (4715169a40)","roles":["extraction"],"url":"https://sah.borca.ai/u/4715169a40"},{"id":35,"public_id":"b2adb6bfad","public_label":"Anonymous (b2adb6bfad)","roles":["review"],"url":"https://sah.borca.ai/u/b2adb6bfad"},{"id":136,"public_id":"3c2apqe3ut","public_label":"Anonymous (3c2apqe3ut)","roles":["review"],"url":"https://sah.borca.ai/u/3c2apqe3ut"}],"url":"https://sah.borca.ai/claims/cl_33d25f5bda751b8f891625fd16ee10d8"}],"concepts":[{"public_id":"co_631a80b41a534a41834d3f249b11f811","status":"active","name":"feed forward neural network","description":"A baseline neural network architecture used for comparison in Korean speech recognition acoustic modeling.","types":["baseline","model"],"aliases":["FFNN"],"contributors":[{"id":2,"public_id":"4715169a40","public_label":"AK (4715169a40)","roles":["extraction"],"url":"https://sah.borca.ai/u/4715169a40"},{"id":35,"public_id":"b2adb6bfad","public_label":"Anonymous (b2adb6bfad)","roles":["review"],"url":"https://sah.borca.ai/u/b2adb6bfad"},{"id":136,"public_id":"3c2apqe3ut","public_label":"Anonymous (3c2apqe3ut)","roles":["review"],"url":"https://sah.borca.ai/u/3c2apqe3ut"}],"url":"https://sah.borca.ai/concepts/co_631a80b41a534a41834d3f249b11f811"},{"public_id":"co_65a55cdd1020e3c140a78a1b864decb2","status":"active","name":"acoustic modeling","description":"The task of modeling speech acoustics for Korean speech recognition, performed here using TDNN and FFNN.","types":["task"],"aliases":[],"contributors":[{"id":2,"public_id":"4715169a40","public_label":"AK (4715169a40)","roles":["extraction"],"url":"https://sah.borca.ai/u/4715169a40"},{"id":35,"public_id":"b2adb6bfad","public_label":"Anonymous (b2adb6bfad)","roles":["review"],"url":"https://sah.borca.ai/u/b2adb6bfad"},{"id":136,"public_id":"3c2apqe3ut","public_label":"Anonymous (3c2apqe3ut)","roles":["review"],"url":"https://sah.borca.ai/u/3c2apqe3ut"}],"url":"https://sah.borca.ai/concepts/co_65a55cdd1020e3c140a78a1b864decb2"},{"public_id":"co_701afa6e77176edf5ec7d346a40f2b5d","status":"active","name":"time delay neural network","description":"A neural network architecture that uses subsampling to exclude duplicated weights, enabling fast convergence with limited training data.","types":["method","model"],"aliases":["TDNN"],"contributors":[{"id":2,"public_id":"4715169a40","public_label":"AK (4715169a40)","roles":["extraction"],"url":"https://sah.borca.ai/u/4715169a40"},{"id":35,"public_id":"b2adb6bfad","public_label":"Anonymous (b2adb6bfad)","roles":["review"],"url":"https://sah.borca.ai/u/b2adb6bfad"},{"id":136,"public_id":"3c2apqe3ut","public_label":"Anonymous (3c2apqe3ut)","roles":["review"],"url":"https://sah.borca.ai/u/3c2apqe3ut"}],"url":"https://sah.borca.ai/concepts/co_701afa6e77176edf5ec7d346a40f2b5d"},{"public_id":"co_7b8be8194ac21be92fc12ab1b73ff4d2","status":"active","name":"subsampling","description":"The technique in TDNN that excludes duplicated weights to enable faster convergence with limited training data.","types":["technique"],"aliases":[],"contributors":[{"id":2,"public_id":"4715169a40","public_label":"AK (4715169a40)","roles":["extraction"],"url":"https://sah.borca.ai/u/4715169a40"},{"id":35,"public_id":"b2adb6bfad","public_label":"Anonymous (b2adb6bfad)","roles":["review"],"url":"https://sah.borca.ai/u/b2adb6bfad"},{"id":136,"public_id":"3c2apqe3ut","public_label":"Anonymous (3c2apqe3ut)","roles":["review"],"url":"https://sah.borca.ai/u/3c2apqe3ut"}],"url":"https://sah.borca.ai/concepts/co_7b8be8194ac21be92fc12ab1b73ff4d2"},{"public_id":"co_838cd1fda3d8e0350a9d6264e4d7faeb","status":"active","name":"convergence speed","description":"The relative training speed improvement factor (1.67 times faster) achieved by the TDNN compared to the FFNN.","types":["property"],"aliases":[],"contributors":[{"id":2,"public_id":"4715169a40","public_label":"AK (4715169a40)","roles":["extraction"],"url":"https://sah.borca.ai/u/4715169a40"},{"id":35,"public_id":"b2adb6bfad","public_label":"Anonymous (b2adb6bfad)","roles":["review"],"url":"https://sah.borca.ai/u/b2adb6bfad"},{"id":136,"public_id":"3c2apqe3ut","public_label":"Anonymous (3c2apqe3ut)","roles":["review"],"url":"https://sah.borca.ai/u/3c2apqe3ut"}],"url":"https://sah.borca.ai/concepts/co_838cd1fda3d8e0350a9d6264e4d7faeb"},{"public_id":"co_8e266938e0942a67944ccf5824e0b68d","status":"active","name":"Korean speech corpora","description":"The speech datasets used for training and evaluating the acoustic models in this study.","types":["dataset"],"aliases":[],"contributors":[{"id":2,"public_id":"4715169a40","public_label":"AK (4715169a40)","roles":["extraction"],"url":"https://sah.borca.ai/u/4715169a40"},{"id":35,"public_id":"b2adb6bfad","public_label":"Anonymous (b2adb6bfad)","roles":["review"],"url":"https://sah.borca.ai/u/b2adb6bfad"},{"id":136,"public_id":"3c2apqe3ut","public_label":"Anonymous (3c2apqe3ut)","roles":["review"],"url":"https://sah.borca.ai/u/3c2apqe3ut"}],"url":"https://sah.borca.ai/concepts/co_8e266938e0942a67944ccf5824e0b68d"},{"public_id":"co_8f6b0a0ab3dab3557c44a46b3c2c0cd8","status":"active","name":"character error rate","description":"The evaluation metric used to measure the performance improvement of the TDNN over the FFNN in Korean speech recognition.","types":["metric"],"aliases":["CER"],"contributors":[{"id":2,"public_id":"4715169a40","public_label":"AK (4715169a40)","roles":["extraction"],"url":"https://sah.borca.ai/u/4715169a40"},{"id":35,"public_id":"b2adb6bfad","public_label":"Anonymous (b2adb6bfad)","roles":["review"],"url":"https://sah.borca.ai/u/b2adb6bfad"},{"id":136,"public_id":"3c2apqe3ut","public_label":"Anonymous (3c2apqe3ut)","roles":["review"],"url":"https://sah.borca.ai/u/3c2apqe3ut"}],"url":"https://sah.borca.ai/concepts/co_8f6b0a0ab3dab3557c44a46b3c2c0cd8"},{"public_id":"co_af669cfa564b725da2005fb4dba7445a","status":"active","name":"fast convergence","description":"The advantage of TDNN in achieving quicker training convergence when the amount of training data is limited.","types":["property"],"aliases":[],"contributors":[{"id":2,"public_id":"4715169a40","public_label":"AK (4715169a40)","roles":["extraction"],"url":"https://sah.borca.ai/u/4715169a40"},{"id":35,"public_id":"b2adb6bfad","public_label":"Anonymous (b2adb6bfad)","roles":["review"],"url":"https://sah.borca.ai/u/b2adb6bfad"},{"id":136,"public_id":"3c2apqe3ut","public_label":"Anonymous (3c2apqe3ut)","roles":["review"],"url":"https://sah.borca.ai/u/3c2apqe3ut"}],"url":"https://sah.borca.ai/concepts/co_af669cfa564b725da2005fb4dba7445a"}],"external_ids":{"DOI":null,"ArXiv":"1807.05855","PubMed":null,"PubMedCentral":null,"MAG":2883762569,"DBLP":"journals/corr/abs-1807-05855","ACL":null},"open_access":{"is_open_access":true,"pdf_url":"https://arxiv.org/pdf/1807.05855","landing_url":"https://arxiv.org/abs/1807.05855","source":"arxiv","pdf_url_source":"derived_arxiv","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","reason":null},"reference_availability":{"status":"available","references_indexed":true,"full_text_available":true,"full_text_source":"arxiv","count_basis":"semantic_scholar_metadata","extraction_status":"not_applicable","reason":null},"source":{"provider":"episteme2","base_corpus":"semantic_scholar_dump","freshness_mode":"unknown","basis":["semantic_scholar_metadata","postgres_metadata"],"limits":["paper metadata is based on indexed upstream scholarly datasets","claims and concepts are available only for extracted papers","absence of claims or concepts means no extracted graph data is available in this response"],"status":"available","degraded":false,"degraded_reasons":[],"diagnostics":{"status":"available","degraded":false,"degraded_reasons":[],"metadata_status":"available","graph_status":"available","abstract_status":"available"},"source_flags":1},"paper_id":632916,"paper_uid":"593b2a0b-14cb-4ba4-8ff7-bd36994d02f1","canonical_identity":{"paper_id":632916,"paper_uid":"593b2a0b-14cb-4ba4-8ff7-bd36994d02f1","identity_status":"available","lookup_basis":"semantic_scholar_external_id","compatibility_path":"corpus_id"},"url":"https://sah.borca.ai/papers/49862364"}