Compressed sensing based speech enhancement

Firouzeh, Fereshteh Fakhar; Ghorshi, Seyed; Salsabili, Sina

doi:10.1109/icspcs.2014.7021068

Cited by 9 publications

(6 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Fig. 6 Influence of balance factor β on performance BP algorithm [24] is selected to obtain the sparse representation in this paper.…”

Section: Joint Constrained Dictionary Learningmentioning

confidence: 99%

Single-channel speech enhancement based on joint constrained dictionary learning

Sun

et al. 2021

J AUDIO SPEECH MUSIC PROC.

View full text Add to dashboard Cite

To improve the performance of speech enhancement in a complex noise environment, a joint constrained dictionary learning method for single-channel speech enhancement is proposed, which solves the “cross projection” problem of signals in the joint dictionary. In the method, the new optimization function not only constrains the sparse representation of the noisy signal in the joint dictionary, and controls the projection error of the speech signal and noise signal on the corresponding sub-dictionary, but also minimizes the cross projection error and the correlation between the sub-dictionaries. In addition, the adjustment factors are introduced to balance the weight of constraint terms to obtain the joint dictionary more discriminatively. When the method is applied to the single-channel speech enhancement, speech components of the noisy signal can be more projected onto the clean speech sub-dictionary of the joint dictionary without being affected by the noise sub-dictionary, which makes the quality and intelligibility of the enhanced speech higher. The experimental results verify that our algorithm has better performance than the speech enhancement algorithm based on discriminative dictionary learning under white noise and colored noise environments in time domain waveform, spectrogram, global signal-to-noise ratio, subjective evaluation of speech quality, and logarithmic spectrum distance.

show abstract

“…Fig. 6 Influence of balance factor β on performance BP algorithm [24] is selected to obtain the sparse representation in this paper.…”

Section: Joint Constrained Dictionary Learningmentioning

confidence: 99%

Single-channel speech enhancement based on joint constrained dictionary learning

Sun

et al. 2021

J AUDIO SPEECH MUSIC PROC.

View full text Add to dashboard Cite

show abstract

“…Several methods have been proposed in the last few decades to solve the problem of noise reduction and speech enhancement. Among these methods, time domain methods such as subspace approach [11,15], frequency domain methods like discrete cosine transform based methods [6], spectral subtraction [5,17,18,39], minimum mean square error (MMSE) estimator [11,25], Weiner filtering [2,4], compressive sensing based methods [13,37,38] and timefrequency domain methods such as wavelet or wavelet packet based thresholding methods [3,10,19,23,35] are the prominent ones. These methods have their own advantages and disadvantages.…”

Section: Introductionmentioning

confidence: 99%

“…MMSE and Wiener filtering based methods have moderate computational load but o er no mechanism to control the trade-o between speech distortion and noise reduction. Compressive sensing based methods applied in frequency domain are recent methods which have shown very promising results in reduction of noise while preserving the speech [13,37,38].…”

Section: Introductionmentioning

confidence: 99%

“…methods like discrete cosine transform based methods [6], spectral subtraction [5,17,18,39], minimum mean square error (MMSE) estimator [11,25], Weiner filtering [2,4], compressive sensing based methods [13,37,38] and timefrequency domain methods such as wavelet or wavelet packet based thresholding methods [3,10,19,23,35] are the prominent ones. These methods have their own advantages and disadvantages.…”

mentioning

confidence: 99%

See 1 more Smart Citation

Rayleigh modeling of teager energy operated perceptual wavelet packet coefficients for enhancing noisy speech

Islam

Shahnaz

Zhu

et al. 2017

Speech Communication

View full text Add to dashboard Cite

“…In [184], the superior performance of SL0 when compared with FOCUSS and BP (solved by interior-point LP solvers) was demonstrated in a a blind Source Separation (BSS) application. Moreover, [120] presented the results of CoSaMP and BP in speech enhancement and showed that BP outperforms CoSaMP in speech recovery quality.…”

Section: Sl0 Focuss Bpmentioning

confidence: 99%

Sparse Recovery, Classification, and Data Compression via Improved Maximum Feasible Subsystem Algorithms

Firouzeh¹

View full text Add to dashboard Cite

Two new strategies are developed in this thesis to increase the speed of the state-ofthe-art Maximum Feasible Subsystem (MAX FS) algorithms. The newly developed strategies can be combined with any MAX FS algorithm to increase its speed while preserving or improving solution quality. The improved algorithms apply in the case of dense constraint matrices such as those found when various data compression/dimensionality reduction, classification, and sparse recovery problems are converted to MAX FS problems. This approach is used for sparse recovery in Compressive Sensing (CS) and data compression via Nonnegative Matrix Factorization (NNMF) for the first time in this thesis. In CS, the new algorithm can successfully recover real-world signals such as speech and Electrocardiogram (ECG) signals that have been more greatly compressed than existing methods can handle, and with greater recovered signal quality. They also deliver sparser solutions when compared with those obtained using traditional sparse recovery algorithms. The new algorithm significantly reduces the solution time of the existing MAX FS methods. It reduces the solution time of Method B by 80.64% and 77.82% for recovery of compressively sensed ECG and speech signals for instance, respectively. Three novel MAX FS-based methods for NNMF are developed and investigated using synthetic data and real-world image datasets. The new NNMF algorithms, unlike the majority of existing NNMF methods, do not require an initialization method or prior knowledge of the matrix rank. iii These new NNMF solutions methods are also more robust to outliers and provide higher quality solution compared to the state-of-the-art NNMF algorithms. To illustrate, the new NNMF algorithm improves the relative approximation quality of the existing column subset selection NNMF algorithm by 5.2% for the ORL facial image dataset. A faster MAX FS-based algorithm for binary classification is also proposed, which improves the processing time of the existing MAX FS algorithm (Algorithm 2) by 94.77% on average. In comparison, the developed algorithm outperforms widely used binary classification algorithms by providing better results for recall-oriented machine learning tasks such as disease diagnosis. For instance, the new algorithm improves the average recall score of Support Vector Machine (SVM) by 18.16% for binary classification task.

show abstract

Compressed sensing based speech enhancement

Cited by 9 publications

References 25 publications

Single-channel speech enhancement based on joint constrained dictionary learning

Single-channel speech enhancement based on joint constrained dictionary learning

Rayleigh modeling of teager energy operated perceptual wavelet packet coefficients for enhancing noisy speech

Sparse Recovery, Classification, and Data Compression via Improved Maximum Feasible Subsystem Algorithms

Contact Info

Product

Resources

About