Variational Bayesian model averaging for audio source separation

Jaureguiberry, Xabier; Richard, Gaël

doi:10.1109/ssp.2014.6884568

Cited by 4 publications

(5 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In this study, we further extend our preliminary works [26], [29] to a novel adaptive time-varying fusion rule in which the fusion weights are adapted to each frame of the mixture to be separated. As such, the general fusion framework that we propose nicely handles all the previously introduced fusion rules.…”

Section: Introductionmentioning

confidence: 85%

“…The term C j,f n depends on the NMF parameters of source j through the expectation (34) in which E denotes the expectation over the variational posteriors q(w j,f k ) and q(h j,kn ) which turn out to be generalized inverse Gaussian (GIG) distributions. For the detailed expressions of Σ s,f n , q(w j,f k ), q(h j,kn ) and C j,f n , see [29].…”

Section: B Variational Bayesian Formulationmentioning

confidence: 99%

“…Following our previous works [22], [26], [29], we here propose a general framework for fusion in audio source separation which only assumes that the considered separation techniques lead to time-domain estimated signals. This allows the combination of heterogeneous separation techniques as well as identical techniques with different parameter settings.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Fusion Methods for Speech Enhancement and Audio Source Separation

Jaureguiberry¹,

Vincent

Richard

2016

IEEE/ACM Trans. Audio Speech Lang. Process.

Self Cite

View full text Add to dashboard Cite

A wide variety of audio source separation techniques exist and can already tackle many challenging industrial issues. However, in contrast with other application domains, fusion principles were rarely investigated in audio source separation despite their demonstrated potential in classification tasks. In this paper, we propose a general fusion framework which takes advantage of the diversity of existing separation techniques in order to improve separation quality. We obtain new source estimates by summing the individual estimates given by different separation techniques weighted by a set of fusion coefficients. We investigate three alternative fusion methods which are based on standard non-linear optimization, Bayesian model averaging or deep neural networks. Experiments conducted for both speech enhancement and singing voice extraction demonstrate that all the proposed methods outperform traditional model selection. The use of deep neural networks for the estimation of timevarying coefficients notably leads to large quality improvements, up to 3 dB in terms of signal-to-distortion ratio (SDR) compared to model selection.

show abstract

Section: Introductionmentioning

confidence: 85%

Section: B Variational Bayesian Formulationmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Fusion Methods for Speech Enhancement and Audio Source Separation

Jaureguiberry¹,

Vincent

Richard

2016

IEEE/ACM Trans. Audio Speech Lang. Process.

Self Cite

View full text Add to dashboard Cite

show abstract

“…Our novel multiple-order NMF model introduced here aims at jointly estimating and averaging several NMFs of different orders. For more details on the model and the derivation of the VB inference, the reader is referred to [24].…”

Section: Multiple-order Nmfmentioning

confidence: 99%

“…As stated in [21,24], it is worth combining NMFs of different orders instead of selecting a unique order as it can improve the separation performance. However, our models, which apply model averaging in a straightforward way, are not achieving that goal.…”

Section: Controlling the Order Posterior Entropymentioning

confidence: 99%

Multiple-order non-negative matrix factorization for speech enhancement

Jaureguiberry

Richard

2014

Interspeech 2014

Self Cite

View full text Add to dashboard Cite

Amongst the speech enhancement techniques, statistical models based on Non-negative Matrix Factorization (NMF) have received great attention. In a single channel configuration, NMF is used to describe the spectral content of both the speech and noise sources. As the number of components can have a crucial influence on separation quality, we here propose to investigate model order selection based on the variational Bayesian approximation to the marginal likelihood of models of different orders. To go further, we propose to use model averaging to combine several single-order NMFs and we show that a straightforward application of model averaging principles is inefficient as it turned out to be equivalent to model selection. We thus introduce a parameter to control the entropy of the model order distribution which makes the averaging effective. We also show that our probabilistic model nicely extends to a multipleorder NMF model where several NMFs are jointly estimated and averaged. Experiments are conducted on real data from the CHiME challenge and give an interesting insight on the entropic parameter and model order priors. Separation results are also promising as model averaging outperforms single-order model selection. Finally, our multiple-order NMF shows an interesting gain in computation time.

show abstract