2017/08/07 by Nikolaos Dionelis, Dionelis, Nikolaos, Mike Brookes +1
Computer Science · Engineering · #Speech and Audio Processing #Advanced Adaptive Filtering Techniques #Blind Source Separation Techniques
paper · pdf · doi:10.48550/arxiv.1708.02171
We present a single-channel phase-sensitive speech enhancement algorithm that\nis based on modulation-domain Kalman filtering and on tracking the speech phase\nusing circular statistics. With Kalman filtering, using that speech and noise\nare additive in the complex STFT domain, the algorithm tracks the speech\nlog-spectrum, the noise log-spectrum and the speech phase. Joint amplitude and\nphase estimation of speech is performed. Given the noisy speech signal,\nconventional algorithms use the noisy phase for signal reconstruction\napproximating the speech phase with the noisy phase. In the proposed Kalman\nfiltering algorithm, the speech phase posterior is used to create an enhanced\nspeech phase spectrum for signal reconstruction. The Kalman filter prediction\nmodels the temporal/inter-frame correlation of the speech and noise log-spectra\nand of the speech phase, while the Kalman filter update models their nonlinear\nrelations. With the proposed algorithm, speech is tracked and estimated both in\nthe log-spectral and spectral phase domains. The algorithm is evaluated in\nterms of speech quality and different algorithm configurations, dependent on\nthe signal model, are compared in different noise types. Experimental results\nshow that the proposed algorithm outperforms traditional enhancement algorithms\nover a range of SNRs for various noise types.\n