To improve the separation performance of ICA algorithm, wavelet packets transformation was adopted to reduce the signals' overlapped degree, that was, the mixture speech signals were decomposed into wavelet packets, and the node that had the highest kurtosis was the optimal wavelet packets decomposition node since the kurtosis is a measure of non-Gaussian nature. Thereby, it reduced the signals' overlapped degree in the wavelet domain. Then the separation matrix was calculated by using FastICA algorithm iteratively, and the source signal estimations were obtained finally. Simulation results demonstrated the separation performance improved clearly when compared with FastICA algorithm in time domain and other wavelet FastICA method. © Springer International Publishing Switzerland 2014.
CITATION STYLE
Liu, J., Yu, F., & Chen, Y. (2014). Speech separation based on improved fast ICA with kurtosis maximization of wavelet packet coefficients. In Advances in Intelligent Systems and Computing (Vol. 275 AISC, pp. 43–50). Springer Verlag. https://doi.org/10.1007/978-3-319-05951-8_5
Mendeley helps you to discover research relevant for your work.