Enhancing Speech Quality Using Artificial Bandwidth Expansion with Deep Shallow Convolution Neural Network Framework

Radha, N.; Jananie, R. B.; Silviya, A. Anto

doi:10.1142/s0219477522500080

Cited by 3 publications

(1 citation statement)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As one of the core technologies in information science, speech enhancement technology is mainly used to enhance the clarity and intelligibility of speech polluted by noise. In recent years, speech enhancement has been widely used in mobile phones, smart appliances, multimedia applications, military eavesdropping, and other fields [1][2][3][4]. At the same time, speech recognition technology also needs speech enhancement technology [5].…”

Section: Introductionmentioning

confidence: 99%

Research on Speech Enhancement Algorithm of Multiresolution Cochleagram Based on Skip Connection Deep Neural Network

Lan

Wang

Zhang³

et al. 2022

Journal of Sensors

View full text Add to dashboard Cite

The speech enhancement effect of traditional deep learning algorithms is not ideal under low signal-to-noise ratios (SNR). Skip connections-deep neural network (Skip-DNN) improves the traditional deep neural network (DNN) by adding skip connections between each layer of the neural network to solve the degradation problem of DNN. In this paper, the Multiresolution Cochleagram (MRCG) features in the gammachirp transform domain are denoised to obtain the improved MRCG (I-MRCG). The noise reduction method adopts the Minimum Mean-Square Error Short-Time Spectral Amplitude Estimator (MMSE-STSA) and takes I-MRCG as the input feature and Skip-DNN as the training network to improve the speech enhancement effect of the model. This paper also proposes an improved source-to-distortion ratio (SDR) loss function. When the loss function uses the improved SDR, it will improve the performance of Skip-DNN speech enhancement model. The experiments in this paper are performed on the Edinburgh dataset. When using I-MRCG as the input feature of Skip-DNN, the average perceptual evaluation of speech quality (PESQ) is 2.9137, and the average short-time objective intelligibility (STOI) is 0.8515. Compared with MRCG as Skip-DNN input features, the improvements are 0.91% and 0.71%, respectively. When the improved SDR is used as the loss function of the speech model, the average PESQ is 2.9699 and the average STOI is 0.8547. Compared with other loss functions, the improved SDR has a better enhancement effect when used as the loss function of the speech enhancement model.

show abstract

Section: Introductionmentioning

confidence: 99%

Research on Speech Enhancement Algorithm of Multiresolution Cochleagram Based on Skip Connection Deep Neural Network

Lan

Wang

Zhang³

et al. 2022

Journal of Sensors

View full text Add to dashboard Cite

show abstract

Fused Lasso Algorithm Based on Novel Non-Convex Regularization in Sparse Domain for Audio Signal Enhancement

Kittisuwan¹,

Akkaraekthalin²

2023

Fluct. Noise Lett.

View full text Add to dashboard Cite

In this paper, we present the fused lasso algorithm in the wavelet domain for audio signal enhancement in this research. The fused lasso algorithm gives more efficiency than other methods because this algorithm comprises two parts. The first part is called as the total variation method (TVM), and the second part is known as the penalized least squares regression (PLSR). The prominent point of the TVM is based on decreasing the pseudo-Gibbs phenomena. On the other hand, the PLSR with the convex condition, also called as the convex PLSR, gives the advantage in reducing noise. A lot of works study and develop the TVM; therefore, we focus on developing the convex PLSR in this work. In fact, the bias which is the gap between the estimated and noisy signals at the large magnitude of the noisy signal affects the performance of the convex PLSR. If this bias is small, also known as the small bias, then the convex PLSR gives better efficiency than other forms. Therefore, we present the novel non-convex regularization which can build the convex PLSR in the closed-form solution and the small bias. Here, the convex PLSR of the proposed regularization is presented not only in the closed-form solution and the small bias but also in the relationship to the group of wavelet coefficients, also called as the multivariate case. In the signal enhancement problem, experimental results show that the fused lasso algorithm and the convex PLSR based on the proposed regularization outperform state-of-the-art techniques in both the error of the [Formula: see text] norm (L2E) and visual quality.

show abstract

Constructing Signal from Imperfect Data without Prior Information and Training Data

Kittisuwan,

Woradit

2024

Fluct. Noise Lett.

View full text Add to dashboard Cite

In many situations, the signal can be disrupted not only by noise but also by missing data. Many works present deep learning techniques to solve the noise and missing-data problems. These techniques give good efficiency for removing adulterated things. However, many deep learning techniques do not give good efficiency in computational time because these methods require large architecture and training data via prior information. Moreover, prior information may be lacking in some situations. Therefore, we present the algorithm which comprises the least squares estimation to solve the missing-data problem and the penalized least squares regression (PLSR) for solving the noise problem. The proposed algorithm is based on the novel channel model which is not similar to the traditional form of additive white Gaussian noise (AWGN). Here, the proposed method is not similar to deep learning methods, so this method gives very good efficiency in computational time. Moreover, the proposed shrinkage function which is built by the PLSR is the generalized form of the well-known shrinkage function presented in the past. Finally, we use the speech, electrocardiogram (ECG) and synthetic signals corrupted by adulterated things for testing the proposed algorithm. Experimental results show that the proposed method gives good efficiency for creating data.

show abstract

Enhancing Speech Quality Using Artificial Bandwidth Expansion with Deep Shallow Convolution Neural Network Framework

Cited by 3 publications

References 29 publications

Research on Speech Enhancement Algorithm of Multiresolution Cochleagram Based on Skip Connection Deep Neural Network

Research on Speech Enhancement Algorithm of Multiresolution Cochleagram Based on Skip Connection Deep Neural Network

Fused Lasso Algorithm Based on Novel Non-Convex Regularization in Sparse Domain for Audio Signal Enhancement

Constructing Signal from Imperfect Data without Prior Information and Training Data

Contact Info

Product

Resources

About