REPORT Cernak_Idiap-RR-02-2013/IDIAP Robust triphone mapping for acoustic modeling Cernak, Milos Imseng, David Bourlard, Hervé EXTERNAL https://publications.idiap.ch/attachments/reports/2012/Cernak_Idiap-RR-02-2013.pdf PUBLIC https://publications.idiap.ch/index.php/publications/showcite/Cernak_INTERSPEECH_2012 Related documents Idiap-RR-02-2013 2013 Idiap January 2013 In this paper we revisit the recently proposed triphone mapping as an alternative to decision tree state clustering. We generalize triphone mapping to Kullback-Leibler based hidden Markov models for acoustic modeling and propose a modified training procedure for the Gaussian mixture model based acoustic modeling. We compare the triphone mapping to decision tree state clustering on the Wall Street journal task as well as in the context of an under-resourced language by using Greek data from the SpeechDat(II) corpus. Experiments reveal that triphone mapping has the best overall performance and is robust against varying the acoustic modeling technique as well as variable amounts of training data. CONF Cernak_INTERSPEECH_2012/IDIAP Robust triphone mapping for acoustic modeling Cernak, Milos Imseng, David Bourlard, Hervé acoustic modeling Kullback-Leibler divergence speech recognition triphone mapping EXTERNAL https://publications.idiap.ch/attachments/papers/2012/Cernak_INTERSPEECH_2012.pdf PUBLIC Proceedings of Interspeech Portland, Oregon 2012 In this paper we revisit the recently proposed triphone mapping as an alternative to decision tree state clustering. We generalize triphone mapping to Kullback-Leibler based hidden Markov models for acoustic modeling and propose a modified training procedure for the Gaussian mixture model based acoustic modeling. We compare the triphone mapping to decision tree state clustering on the Wall Street journal task as well as in the context of an under-resourced language by using Greek data from the SpeechDat(II) corpus. Experiments reveal that triphone mapping has the best overall performance and is robust against varying the acoustic modeling technique as well as variable amounts of training data.