It is well known that the performances of speaker identification systems degrade rapidly as the mismatch between training and test conditions increases. In this work we present a noise compensation technique whose goal is to minimize the effects of such mismatch, so as to obtain an identification accuracy as close as possible to that obtained under matched conditions. To reduce this mismatch, the adopted approach compensates the speaker model parameters using the noise present in the test data, and compensates the test data frames using the noise present in the training data. The test and the training data (for different speakers) are assumed to come from different and unknown microphones and acoustic environments.
Cite as: Bellot, O., Matrouf, D., Merlin, T., Bonastre, J.-F. (2000) Additive and convolutional noises compensation for speaker recognition. Proc. 6th International Conference on Spoken Language Processing (ICSLP 2000), vol. 2, 799-802, doi: 10.21437/ICSLP.2000-390
@inproceedings{bellot00_icslp,
author={Olivier Bellot and Driss Matrouf and Teva Merlin and Jean-François Bonastre},
title={{Additive and convolutional noises compensation for speaker recognition}},
year=2000,
booktitle={Proc. 6th International Conference on Spoken Language Processing (ICSLP 2000)},
pages={vol. 2, 799-802},
doi={10.21437/ICSLP.2000-390}
}