Abstract:
In the field of speaker recognition, the noise of the speech interfered with the correct rate of speaker recognition largely. In order to filter out the noise in the noisy speech and further to improve the correct rate of speaker recognition, a weighted mathematical morphological filter was proposed in this study which can denoise the one-dimensional noisy speech signal. The morphological pre-filtering was conducted before training the speech to improve the matching abilities between the speech to be recognized and the training template. Experiment results showed that, compared with conventional spectral subtraction method, the method presented in this study suppressed the noise effectively and the signal to noise ratio of the output speech was significantly improved and the error rate of the speaker recognition system was further reduced.
Subscribe Today 
