ISCA Archive Interspeech 2013
ISCA Archive Interspeech 2013

Noise adaptive training for subspace Gaussian mixture models

Liang Lu, Arnab Ghoshal, Steve Renals

Noise adaptive training (NAT) is an effective approach to normalise environmental distortions when training a speech recogniser on noise-corrupted speech. This paper investigates the model-based NAT scheme using joint uncertainty decoding (JUD) for subspace Gaussian mixture models (SGMMs). A typical SGMM acoustic model has much larger number of surface Gaussian components, which makes it computationally infeasible to compensate each Gaussian explicitly. JUD tackles this problem by sharing the compensation parameters among the Gaussians and hence reduces the computational and memory demands. For noise adaptive training, JUD is reformulated into a generative model, which leads to an efficient expectation-maximisation (EM) based algorithm to update the SGMM acoustic model parameters. We evaluated the SGMMs with NAT on the Aurora 4 database, and obtained higher recognition accuracy compared to systems without adaptive training.


doi: 10.21437/Interspeech.2013-263

Cite as: Lu, L., Ghoshal, A., Renals, S. (2013) Noise adaptive training for subspace Gaussian mixture models. Proc. Interspeech 2013, 3492-3496, doi: 10.21437/Interspeech.2013-263

@inproceedings{lu13b_interspeech,
  author={Liang Lu and Arnab Ghoshal and Steve Renals},
  title={{Noise adaptive training for subspace Gaussian mixture models}},
  year=2013,
  booktitle={Proc. Interspeech 2013},
  pages={3492--3496},
  doi={10.21437/Interspeech.2013-263}
}