Aicha Bouzid

Speech enhancement based on wavelet packet of an improved principal component analysis

Integrating the principal component analysis in wavelet packet decomposition.Extended PCA technique for speech enhancement is considered.To obtain a sparse matrix to contain the enhanced speech.Experiments on NOIZEUS data corrupted by Gaussian and four non-stationary noises.Our approach shows superior outcomes in BSS EVAL toolbox, SegSNR, PESQ, and Cov. In this paper, we propose a single-channel speech enhancement method, based on the combination of the wavelet packet transform and an improved version of the principal component analysis (PCA). Our method integrates ability of PCA to de-correlate the coefficients by extracting a linear relationship with what of wavelet packet analysis to derive feature vectors used for speech enhancement. This allows us to operate with a convenient shrinkage function on these new coefficients, removing the noise without degrading the speech. Then, the enhanced speech obtained by the inverse wavelet packet transform is decomposed into three subspaces: low rank, sparse, and the remainder noise components. Finally, we calculate the components as a segregation problem. The performance evaluation shows that our method provides a higher noise reduction and a lower signal distortion even in highly noisy conditions without introducing artifacts.

A New Biologically Inspired Fuzzy Expert System-Based Voiced/Unvoiced Decision Algorithm for Speech Enhancement

In this paper, we propose a speech enhancement approach for a single-microphone system. The main idea is to apply a specific transformation on the speech signal depending on the voicing state of the signal. We apply a voiced/unvoiced algorithm based on the multi-scale product analysis with the use of fuzzy logic to make more cognitively inspired use of speech information. A comb filtering is applied on the voiced frames of the noisy speech signal, and a spectral subtraction is operated on the unvoiced frames of the same signal. Further, the harmonics are enhanced by performing a designed comb filtering using an adjustable bandwidth. The comb filter is tuned by an accurate fundamental frequency estimation method. The fundamental frequency estimation method is based on computing the multi-scale product analysis of the noisy speech. Experimental results show that the proposed approach is capable of reducing noise in adverse noise environments with little speech degradation and outperforms several competitive methods.

Sparse Representations for Single Channel Speech Enhancement Based on Voiced/Unvoiced Classification

The approach presented here in relies on a new voicing decision algorithm based on the multi-scale product (MP) characteristics. The MP is based on the multiplication of Wavelet Transform Coefficients at some scales. According to the voicing decision, improved subspace decomposition is operated on the voiced segments of the noisy speech signal and a multi-scale principal component analysis is applied on the unvoiced segments of the same signal. Furthermore, the voiced frames are decomposed into three subspaces: sparse, low rank, and the remainder noise components. Then, we calculate the components as a segregation problem. In the unvoiced frames, we combine the straightforward multivariate generalization of the wavelet denoising technique with the principal component analysis method. Experiments on NOIZEUS and NTT databases show that the proposed approach obtains satisfying results for most types of noise with little speech degradation and outperforms several competitive methods.

An Efficient Method for Fundamental Frequency Determination of Noisy Speech

In this paper, we present a fundamental frequency determination method dependent on the autocorrelation compression of the multi-scale product of speech signal. It is based on the multiplication of compressed copies of the original autocorrelation operated on the multi-scale product. The multi-scale product is based on realising the product of the speech wavelet transform coefficients at three successive dyadic scales. We use the quadratic spline wavelet function. We compress the autocorrelation of the multi-scale product a number of times by integer factors (downsampling). Hence, when the obtained functions are multiplied, we obtain a peak with a clear maximum corresponding to the fundamental frequency. We have evaluated our method on the Keele database. Experimental results show the effectiveness of our method presenting a good performance surpassing other algorithms. Besides, the proposed approach is robust in noisy environment.

Multi-pitch estimation based on multi-scale product analysis, improved comb filter and dynamic programming

There are many multi-pitch estimation methods, but most of them can’t perform perfectly for intrusion pitch detection. For this reason, a new multi-pitch detection approach is proposed. This method consists on the autocorrelation function of the Multi-scale product calculation of the mixture signal, its filtered version by a rectangular improved comb filter and the dynamic programming of the residual signal spectral density. First, we analyze the composite speech. Then, we apply the autocorrelation on the multi-scale product (AMP). We find the first pitch which represents the dominant one. Then, we apply the rectangular comb filter which has adaptive amplitude to remove the resulting signal from the original one. We operate AMP on the residue to obtain a pitch estimation of the intrusion. To improve the residue pitch estimation, we apply the dynamic programming to the spectral density of the residual signal to get optimum pitches corresponding also to intrusion signal. After that, we compare the two resulting pitch residue series to choose the most appropriate. Finally, this method is evaluated using the Cooke database and is compared to other well-known techniques. Experimental results confirm the strength and the performance of the proposed approach.

Multiple fundamental frequencies estimation approaches based on multi-scale product analysis

This paper describes three methods for multiple fundamental frequencies estimation based on the multi-scale product analysis. The three methods use the autocorrelation of the multi-scale product analysis for the target pitch estimation. For the intrusion pitch, each one has its techniques. The first one uses the classic comb filtering. The second method employs the rectangular comb filter followed by the dynamic programming and the third one uses multiple-comb filters. These methods are evaluated on the Cooke database by calculating the gross error and the root means square error and compared to each other.

A comparison of several computational auditory scene analysis (CASA) techniques for monaural speech segregation

Humans have the ability to easily separate a composed speech and to form perceptual representations of the constituent sources in an acoustic mixture thanks to their ears. Until recently, researchers attempt to build computer models of high-level functions of the auditory system. The problem of the composed speech segregation is still a very challenging problem for these researchers. In our case, we are interested in approaches that are addressed to the monaural speech segregation. For this purpose, we study in this paper the computational auditory scene analysis (CASA) to segregate speech from monaural mixtures. CASA is the reproduction of the source organization achieved by listeners. It is based on two main stages: segmentation and grouping. In this work, we have presented, and compared several studies that have used CASA for speech separation and recognition.

Contribution to the Multipitch Estimation by Multi-scale Product Analysis

This paper describes a new multipitch estimation method. The proposed approach is based on the calculation of the autocorrelation function of the Multi-scale product of the composite signal and its filtered version by a comb filter. After analyzing the composite speech signal, the autocorrelation applied on the multi-scale product (MP) of the signal allows us to find the first pitch; it’s the dominant one. After applying the comb filter, we substract the resulting signal from the original one. Then we apply the same analysis to the residue to obtain the pitch estimation of the intrusion. Besides, this method is applied and evaluated on the Cooke database. It’s also compared to other well known algorithms. Experimental results show the robustness and the effectiveness of our approach.

Voicing Classification Based on Speech Spectrum Multi-Scale Product

Abstract In this work, we present a new method for voicing detection in speech signals. Our approach is based on the spectral analysis of the speech multi-scale product. It consists of operating a Short Time Fourier Transform (STFT) on the speech multi-scale product. The multi-scale product consists of making the product of the speech wavelet transform coefficients at three successive dyadic scales. We localize all the spectral ray and then we define the groups of maxima positions sorted respecting some rules. Regarding the formed groups a voicing decision is made. We evaluate our approach on the Keele University database. The experimental results show the effectiveness of our method comparatively to the state-of-the-art algorithms. Besides, the proposed approach is robust for noisy speech.