Title :
One microphone speech separaction with deep belief network
Author :
Jie Lin ; Bo Fu ; Jianzhang Chen ; Jie Zheng
Author_Institution :
Sch. of Comput. Sci. & Eng., Univ. of Electron. Sci. & Technol. of China, Chengdu, China
Abstract :
In this paper, we proposed a novel method for speech separation under one-microphone input. The method employs the deep belief networks to build the speech magnitude estimator, which provides a soft mask for extracting the desired speech from the input signal mixed with interference signals. The new approach has been evaluated on mixture speech data and the results demonstrated its efficiency.
Keywords :
microphones; speech processing; deep belief network; desired speech extraction; input signal; interference signals; mixture speech data; one microphone speech separation; one-microphone input; soft mask; speech magnitude estimator; Feature extraction; Hidden Markov models; Probability distribution; Speech; Stochastic processes; Training; Vectors; Speech separation; deep belief network; magnitude estimator;
Conference_Titel :
Wavelet Active Media Technology and Information Processing (ICCWAMTIP), 2013 10th International Computer Conference on
Conference_Location :
Chengdu
Print_ISBN :
978-1-4799-2445-5
DOI :
10.1109/ICCWAMTIP.2013.6716592
