creators_name: Kanevsky, Dimitri creators_name: Sainath, Tara creators_name: Ramabhadran, Bhuvana creators_name: Nahamoo, David creators_id: kanevsky@us.ibm.com creators_id: tsainath@MIT.EDU creators_id: bhuvana@us.ibm.com creators_id: nahamoo@us.ibm.com type: preprint datestamp: 2008-04-30 18:34:51 lastmod: 2011-03-11 08:57:07 metadata_visibility: show title: Generalization of Extended Baum-Welch Parameter Estimation for Discriminative Training and Decoding subjects: comp-sci-speech full_text_status: public keywords: Extended Baum-Welch, optimization, speech abstract: We demonstrate the generalizability of the Extended Baum-Welch (EBW) algorithm not only for HMM parameter estimation but for decoding as well. We show that there can exist a general function associated with the objective function under EBW that reduces to the well-known auxiliary function used in the Baum-Welch algorithm for maximum likelihood estimates. We generalize representation for the updates of model parameters by making use of a differentiable function (such as arithmetic or geometric mean) on the updated and current model parameters and describe their effect on the learning rate during HMM parameter estimation. Improvements on speech recognition tasks are also presented here. date: 2008-04-27 date_type: completed refereed: FALSE referencetext: C.Liu and P.Liu and H.Jiang F.Soong and R. Wang, Constrained Line Search Optimization for Discriminative Training in Speech Recognition, ICASSP, 2007 C.Liu and P.Liu and H.Jiang F.Soong and R. Wang, Constrained Line Search Approach to General Discriminative HMM Training, ASRU 2007 P.S. Gopalakrishnan and D. Kanevsky and D. Nahamoo and A. Nadas, An Inequality for Rational Functions with Applications to Some Statistical Estimation Problems, IEEE Trans. Information Theory, 1991, v.37, n. 1, January D. Kanevsky, Extended Baum Transformations for General Functions, Proc. ICASSP, 2004, D. Kanevsky, Extended Baum Transformations For General Functions, II, Human Language Technologies, IBM, 2005, RC23645(W0506-120) T. N. Sainath and D. Kanevsky and B. Ramabhadran, Gradient Steepness Metrics Using Extended Baum-Welch Transformations for Universal Pattern Recognition Tasks, Proc. ICASSP, April, 2008 T. N. Sainath and V. Zue and D. Kanevsky, Audio Classification using EBW Transformations, Proc. Interspeech, 2007 T. N. Sainath and D. Kanevsky and B. Ramabhadran, Broad Phoentic Recognition in a Hidden Markov Model Framework Using Extended Baum-Welch Transformations, Proc. ASRU, 2007 D. Povey and B.Kingsbury, Evaluation of Proposed Modifications to MPE for Large Scale Discriminative Training, Proc. ICASSP, 2007 D. Povey et. al, Boosted MMI for Model and Feature-space Discriminative Training, Proc. ICASSP, 2008 P.C. Woodland and D. Povey, Large Scale Discriminative Training of Hidden Markov Models for Speech Recognition, Computer Speech and Language, 2002, vol. 16, n. 1, pp. 25-47, January citation: Kanevsky, Dr Dimitri and Sainath, Dr Tara and Ramabhadran, Dr Bhuvana and Nahamoo, Dr David (2008) Generalization of Extended Baum-Welch Parameter Estimation for Discriminative Training and Decoding. [Preprint] document_url: http://cogprints.org/6038/1/main_v5.pdf