ISCA Archive Interspeech 2013
ISCA Archive Interspeech 2013

Towards a more efficient SVM supervector speaker verification system using Gaussian reduction and a tree-structured hash

Richard D. McClanahan, Phillip L. De Leon

Speaker verification (SV) systems that employ maximum a posteriori (MAP) adaptation of a Gaussian mixture model (GMM) universal background model (UBM) incur a significant test-stage com- putational load in the calculation of a posteriori probabilities and sufficient statistics. We propose a multi-layered hash system employing a tree-structured GMM which uses Runnalls' GMM reduction technique. The proposed method is applied only to the test stage and does not require any modifications to the training stage or previously-trained speaker models. With the tree-structured hash system we are able to achieve a factor of 8~ reduction in test-stage computation with no degradation in accuracy. Furthermore, we can achieve computational reductions greater than 21~ with less than 7.5% relative degradation in accuracy.


doi: 10.21437/Interspeech.2013-688

Cite as: McClanahan, R.D., Leon, P.L.D. (2013) Towards a more efficient SVM supervector speaker verification system using Gaussian reduction and a tree-structured hash. Proc. Interspeech 2013, 3670-3673, doi: 10.21437/Interspeech.2013-688

@inproceedings{mcclanahan13_interspeech,
  author={Richard D. McClanahan and Phillip L. De Leon},
  title={{Towards a more efficient SVM supervector speaker verification system using Gaussian reduction and a tree-structured hash}},
  year=2013,
  booktitle={Proc. Interspeech 2013},
  pages={3670--3673},
  doi={10.21437/Interspeech.2013-688}
}