I.-T. Um and M.-H. Kim (Korea)
speaker verification, noise, neural network
A speaker verification system should have robustness to noise for its practical use. Gaussian mixture neural network is designed for speaker verification. Since its output is a posterior probability, it is well suited for hypothesis testing. Therefore, it can produce high recognition rate. When it is used for speaker verification in noisy environment, it is capable of canceling noise to some degree. Some experiments with white noise and colored noise show the results, in which gaussian mixture neural network outperforms gaussian mixture model.
Important Links:
Go Back
