Kleynhans, NMolapo, RDe Wet, Febe2013-01-302013-01-302012-11Kleynhans, N, Molapo, R and De Wet, F. 2012. Acoustic model optimisation for a call routing system. PRASA 2012, CSIR International Convention Centre, Pretoria, 29-30 November 2012978-0-620-54601-0http://www.prasa.org/proceedings/2012/prasa2012-32.pdfhttp://hdl.handle.net/10204/6495PRASA 2012, CSIR International Convention Centre, Pretoria, 29-30 November 2012The paper presents work aimed at optimising acoustic models for the AutoSecretary call routing system. To develop the optimised acoustic models: (1) an appropriate phone set was selected and used to create a pronunciation dictionary, (2) various cepstral normalization techniques were investigated, (3) three South African corpora and multiple training data combinations were used to train the acoustic models, and, (4) model-space transformations were applied. Using an independent testing corpus, which contained proper names and South African language names, a named-language recognition accuracy of 95.11 % and proper name recognition accuracy of 93.31% were obtained.enAutoSecretary call routing systemCall routing systemsAcoustic modelsConference PresentationKleynhans, N., Molapo, R., & De Wet, F. (2012). Acoustic model optimisation for a call routing system. PRASA. http://hdl.handle.net/10204/6495Kleynhans, N, R Molapo, and Febe De Wet. "Acoustic model optimisation for a call routing system." (2012): http://hdl.handle.net/10204/6495Kleynhans N, Molapo R, De Wet F, Acoustic model optimisation for a call routing system; PRASA; 2012. http://hdl.handle.net/10204/6495 .TY - Conference Presentation AU - Kleynhans, N AU - Molapo, R AU - De Wet, Febe AB - The paper presents work aimed at optimising acoustic models for the AutoSecretary call routing system. To develop the optimised acoustic models: (1) an appropriate phone set was selected and used to create a pronunciation dictionary, (2) various cepstral normalization techniques were investigated, (3) three South African corpora and multiple training data combinations were used to train the acoustic models, and, (4) model-space transformations were applied. Using an independent testing corpus, which contained proper names and South African language names, a named-language recognition accuracy of 95.11 % and proper name recognition accuracy of 93.31% were obtained. DA - 2012-11 DB - ResearchSpace DP - CSIR KW - AutoSecretary call routing system KW - Call routing systems KW - Acoustic models LK - https://researchspace.csir.co.za PY - 2012 SM - 978-0-620-54601-0 T1 - Acoustic model optimisation for a call routing system TI - Acoustic model optimisation for a call routing system UR - http://hdl.handle.net/10204/6495 ER -