Academic
Publications
Blind Separation and Dereverberation of Speech Mixtures by Joint Optimization

Blind Separation and Dereverberation of Speech Mixtures by Joint Optimization,10.1109/TASL.2010.2045183,IEEE Transactions on Audio, Speech & Language

Blind Separation and Dereverberation of Speech Mixtures by Joint Optimization   (Citations: 4)
BibTex | RIS | RefWorks Download
Journal: IEEE Transactions on Audio, Speech & Language Processing - TASLP , vol. 19, no. 1, pp. 69-84, 2011
Cumulative Annual
View Publication
The following links allow you to view full publications. These links are maintained by other sources not affiliated with Microsoft Academic Search.
    • ...generalized version of the dereverberation method described in [11]...
    • ...This is described in detail elsewhere [11]...

    Takaaki Horiet al. Real-time meeting recognition and understanding using distant micropho...

    • ...The dereverberation part transforms the eight-channel reverberant spectrum sequences ym(f, t )( m =1 , ··· , 8) received by the microphone array to the same number of dereverberated spectrum sequences xm(f, t). We utilized a generalized version of dereverberation method described in [10]...
    • ...The key to the method is the use of multi-channel weighted linear prediction, which enables us to avoid unduly removing the temporal correlation inherent in speech signals (see [10] for detailed discussion)...

    Shoko Arakiet al. Online meeting recognizer with multichannel speaker diarization

    • ...Therefore, under highly reverberant conditions, we can expect a convolutive mixture model in the time-frequency domain [4] to be a better approximation...
    • ...where ¯ S m is the m-th element of the separated signal vector, given by τ W k,τ Y k,n−τ . The joint pdf fY |Θ(Y |Θ) is the likelihood of the unknown variables Θ := {{θ m }m, {W k,τ }k,τ } given observation Y , which is an objective function for achieving separation and dereverberation in a joint manner, as with [4]...
    • ...later been shown to work successfully for BSS in highly reverberant environments [4]...
    • ...The objective of this paper is to investigate the possibility to improve the performance of this state-of-the-art BSS system [4] by replacing the Gaussian AR source model with the speech model we proposed previously [6], which will be reviewed in the next section...
    • ...τ=1 L(θ) for all k .I f we were able to obtain an estimate of the PSD of each source, namely φ m , we could invoke [4] to perform (S2) and (S3)...
    • ...Baseline1 and Baseline2 refer to Sawada’s method [11] and Yoshioka’s method [4], respectively...

    Hirokazu Kameokaet al. Statistical Model of Speech Signals Based on Composite Autoregressive ...

Sort by: