Title :
State of the art discriminative training of subspace constrained Gaussian mixture models in big training corpora
Author :
Jing Huang ; Olsen, Peder A. ; Goel, Vikas
Author_Institution :
T.J. Watson Res. Center, IBM, Yorktown Heights, NY, USA
Abstract :
Discriminatively trained full-covariance Gaussian mixture models have been shown to outperform its corresponding diagonal-covariance models on large vocabulary speech recognition tasks. However, the size of full-covariance model is much larger than that of diagonal-covariance model and is therefore not practical for use in a real system. In this paper, we present a method to build a large discriminatively trained full-covariance model with large (over 9000 hours) training corpora and still improve performance over the diagonal-covariance model. We then reduce the size of the full-covariance model to the size of its baseline diagonal-covariance model by using subspace constrained Gaussian mixture model (SCGMM). The resulting discriminatively trained SCGMM still retains the performance of its corresponding full-covariance model, and improves 5% relative over the same size diagonal-covariance model on a large vocabulary speech recognition task.
Keywords :
Gaussian processes; covariance analysis; speech recognition; SCGMM; baseline diagonal-covariance model; big training corpora; diagonal-covariance models; discriminative training; full-covariance model; large vocabulary speech recognition tasks; subspace constrained Gaussian mixture models; Data models; Gaussian mixture model; Hidden Markov models; Speech; Speech recognition; Training; Discriminative Training; Full Covariance Modeling; Large Corpora; Subspace Constrained Gaussian Mixture Model;
Conference_Titel :
Acoustics, Speech and Signal Processing (ICASSP), 2013 IEEE International Conference on
Conference_Location :
Vancouver, BC
DOI :
10.1109/ICASSP.2013.6639008