Sparse Representations in Audio and Music: From Coding to Source Separation

Plumbley, Mark D.; Blumensath, Thomas; Daudet, Laurent; Gribonval, Rémi; Davies, Mike E.

doi:10.1109/jproc.2009.2030345

Cited by 175 publications

(102 citation statements)

References 39 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…In this work we will focus on greedy algorithms. These methods have found widespread use in audio processing applications ranging from source separation to audio coding and compression [15], due to their computational efficiency and robust performance. We note that while algorithms based on L1-norm relaxation have received significant attention in the context of signal denoising, greedy methods have been comparatively ignored in this context.…”

Section: Sparse Representationmentioning

confidence: 99%

Sparse denoising of audio by greedy time-frequency shrinkage

Bhattacharya

Depalle

2014

2014 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

View full text Add to dashboard Cite

Matching Pursuit (MP) is a greedy algorithm that iteratively builds a sparse signal representation. This work presents an analysis of MP in the context of audio denoising. By interpreting the algorithm as a simple shrinkage approach, we identify the factors critical to its success, and propose several approaches to improve its performance and robustness. We also develop several model enhancements and introduce an audio denoising approach called Greedy Time-Frequency Shrinkage (GTFS). Numerical experiments are performed on a wide range of audio signals, and we demonstrate that GTFS denoising is able to yield results that are competitive with state-of-the-art audio denoising approaches. Notably, GTFS retains a small percentage of a signal's transform coefficients for building a denoised representation, i.e., it produces very sparse denoised results.ii Résumé L'algorithme de Matching Pursuit (MP) construit par itérations une représentation parcimonieuse du signal, au prix d'un coût de calculélevé. Ce mémoire présente une analyse de l'algorithme de MP dans le contexte du débruitage audio. En interprétant l'algorithme MP comme une méthode de contraction simple (simple shrinkage), nous chercheronsà identifier les facteurs essentielsà son succès, puis proposerons plusieurs approches afin d'en améliorer les performances et la robustesse. Plusieurs améliorations du modèle seront ainsi développées, et une approche du débruitage audio dénommée Greedy Time-Frequency Shrinkage (GTFS) sera présentée en détails. Des expérimentations numériques appliquées a un largeéventail de signaux sonores démontrent que les résultats obtenus par débruitage GTFS s'avèrent compétitifs face aux méthodes de débruitage audio qui constituent l'état de l'art. En particulier, le GTFS ne retient qu'un faible pourcentage des coefficients de la transformée du signal pour en construire une représentation débruitée, et produit ainsi des résultats débruités très compacts.iii

show abstract

Section: Sparse Representationmentioning

confidence: 99%

Sparse denoising of audio by greedy time-frequency shrinkage

Bhattacharya

Depalle

2014

2014 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

View full text Add to dashboard Cite

show abstract

“…This is mainly due to the following three reasons. Firstly, as mentioned earlier, music audio can be made sparser if it is transformed into another domain, such as the TF domain, using an analytically pre-defined dictionary such as discrete Fourier transform (DFT) or discrete cosine transform (DCT) [69] [70]. Recent studies show that signal dictionaries directly adapted from training data using machine learning techniques, based on some optimisation criterion (such as the reconstruction error regularised by a sparsity constraint), can offer better performance than the pre-defined dictionary [71] [72].…”

Section: Future Directionsmentioning

confidence: 99%

Monaural Audio Separation Using Spectral Template and Isolated Note Information

Lal¹,

Wang²

2012

Independent Component Analysis for Audio and Biosignal Applications

View full text Add to dashboard Cite

“…models which describe the magnitude spectra of complex sounds as being composed of a purely additive (no negative components) combinations of spectral atoms, have proven to be adept at separating the target speech from interfering sounds such as noise [1,2], other speakers [3,4], music [5,6,7] and even reverberation [8]. For noise-robust automatic speech recognition (ASR), such compositional models really excel when the atoms also have some temporal extent [9,10].…”

Section: Introductionmentioning

confidence: 99%

Exemplar-based joint channel and noise compensation

Gemmeke

Virtanen

Demuynck

2013

2013 IEEE International Conference on Acoustics, Speech and Signal Processing

View full text Add to dashboard Cite

In this paper two models for channel estimation in exemplar-based noise robust speech recognition are proposed. Building on a compositional model that models noisy speech and a combination of noise and speech atoms, the first model iteratively estimates a filter to best compensate the mismatch with the observed noisy speech. The second model estimates separate filters for the noise and speech atoms. We show that both models enable noise-robust ASR even if the channel characteristics of the noisy speech do not match those of the exemplars in the dictionary. Moreover, the second model, which is able to estimate separate filters for speech and noise, is shown to be robust even in the presence of bandwidth-limited sources.

show abstract

Sparse Representations in Audio and Music: From Coding to Source Separation

Cited by 175 publications

References 39 publications

Sparse denoising of audio by greedy time-frequency shrinkage

Sparse denoising of audio by greedy time-frequency shrinkage

Monaural Audio Separation Using Spectral Template and Isolated Note Information

Exemplar-based joint channel and noise compensation

Contact Info

Product

Resources

About