Comparison of several compensation techniques for robust speaker verification
The performance of the speaker recognition systems degrades rapidly as the mismatch between the training and test conditions increases. In actual telephone-based speaker verification applications, both additive and convolutional noise are present and they influence the error rate considerably. The aim of this paper is to expose and compare a series of experiments with two stochastic compensation techniques which make a speaker verification system more robust against noise. The proposed stochastic compensation techniques proved to be effective, but they do not outperform the cepstral mean subtraction method. Experiments comparing the unnormalized scores versus normalized log-likelihood scores are reported. The results show that the normalized scoring method provides better performance. Another interesting point is the selection and design of the anti-speaker model to be used in the normalized method. Some of the proposed techniques have already been successfully applied in robust speech recognition, and the presented preliminary results proved to be encouraging for robust speaker verification. For the entire collection see [Zbl 0927.00044].
- Blind stochastic feature transformation for channel robust speaker verification
- Speaker compensation in automatic speech recognition
- An integrated study of speaker normalisation and HMM adaptation for noise robust speaker-independent speech recognition
- scientific article; zbMATH DE number 2085387
- Robust speaker recognition: using channel space mapping
- Stochastic feature transformation with divergence-based out-of-handset rejection for robust speaker verification
- Blind stochastic feature transformation for channel robust speaker verification
- Speaker compensation in automatic speech recognition
- scientific article; zbMATH DE number 2089625 (Why is no real title available?)
- scientific article; zbMATH DE number 2089640 (Why is no real title available?)
- Robust speaker recognition: using channel space mapping
- Robust speaker verification with state duration modeling
- Compensation of channel and noise distortions combining normalization and speech enhancement techniques
This page was built for publication: Comparison of several compensation techniques for robust speaker verification
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4942736)
