Speech enhancement based on adaptive wavelet denoising on multitaper spectrum

A reliable estimator for speech presence probability (SPP) can significantly improve the performance of many speech enhancement algorithms. Previous work showed that a good SPP estimator can be obtained by using a smooth a-posteriori signal to noise ratio (SNR) function, which can be achieved by reducing the noise variance when estimating the speech power spectrum. In this paper, a wavelet based denoising algorithm is proposed for such purpose. We first apply the wavelet transform to the periodogram of a noisy speech signal to generate an oracle for indicating the locations of the noise floor in the periodogram. We then make use of that oracle to selectively remove the wavelet coefficients of the noise floor in the log multitaper spectrum (MTS) of the noisy speech. The remaining wavelet coefficients are then used to reconstruct a denoised MTS and in turn generate a smooth a-posteriori SNR function. Simulation results show that the new SPP estimator outperforms the traditional approaches and enables a significantly improvement in the quality and intelligibility of the enhanced speeches. © 2012 IEEE.

It is well known that the "musical noise" encountered in most frequency domain speech enhancement algorithms is partially due to the large variance estimates of the spectra. To address this issue, we propose in this paper the use of low-variance spectral estimators based on wavelet thresholding the multitaper spectra for speech enhancement. A short-time spectral amplitude estimator is derived which incorporates the wavelet-thresholded multitaper spectra. Listening tests showed that the use of multitaper spectrum estimation combined with wavelet thresholding suppressed the musical noise and yielded better quality than the subspace and MMSE algorithms.

In this paper, a new speech enhancement algorithm using the L 1 regularization method in the cepstral domain is proposed. Since voiced speeches have a quasi-periodic nature that allows them to be compactly represented in the cepstral domain, the L 1 regularization technique can be applied to better control the optimization process required in speech enhancement applications. The proposed algorithm starts with the traditional temporal cepstral smoothing (TCS) method which gives the initial estimation of the power spectrum of the clean speech. It is then refined using a modified L 1 regularizer which imposes further constraint to the penalty function based on the feature of speech signals in the cepstral domain. A notable improvement of the proposed algorithm over the traditional method is its adaptability to the non-stationary noise. Performance of the proposed algorithm is evaluated using standard measures such as segSNR and PESQ based on a large quantity of speech signals. Our results show that a significant improvement is achieved as compared to the conventional approaches especially in the case that the noise is non-stationary.

It was suggested that spectrum estimation can be accomplished by applying wavelet denoising methodology to wavelet packet coefficients derived from the logarithm of a spectrum estimate. The particular algorithm we consider consists of computing the logarithm of the multitaper spectrum estimator, applying an orthonormal transform derived from a wavelet packet tree to the log multitaper spectrum ordinates, thresholding the empirical wavelet packet coefficients, and then inverting the transform. For a small number of tapers, suitable transforms/partitions for the logarithm of the multitaper spectrum estimator are derived using a method matched to statistical thresholding properties. The partitions thus derived starting from different stationary time series are all similar and easily derived, and any differences between the wavelet packet and discrete wavelet transform (DWT) approaches are minimal. For a larger number of tapers, where the chosen parameters satisfy the conditions of a proven theorem, the simple DWT again emerges as appropriate. Hence, using our approach to thresholding and the method of partitioning, we conclude that the DWT approach is a very adequate wavelet-based approach and that the use of wavelet packets is unnecessary.

Two gain forms of spectral amplitude subtraction are derived theoretically without neglecting the correlation of speech and noise spectrum during the period of a frame. In the implementation, the constrained gain is expressed as a function of noncausal a priori SNR (Signal-to-Noise Ratio). Noise and noncausal a priori SNR are estimated from the multitaper spectrum of the noisy signal with algorithms modified to be suitable for the multitaper spectrum. Objective evaluations show that in case of white Gaussian noise the proposed method outperforms some methods based on LSA (Log Spectral Amplitude) in terms of MBSD (Modified Bark Spectral Distortion), segmental SNR and overall SNR, and informal listening tests show that speech reconstructed in this way has little speech distortion and musical noise is nearly inaudible even at low SNR.

This paper describes a method for enhancing speech corrupted by broadband noise. The method is based on the spectral noise subtraction method. The original method entails subtracting an estimate of the noise power spectrum from the speech power spectrum, setting negative differences to zero, recombining the new power spectrum with the original phase, and then reconstructing the time waveform. While this method reduces the broadband noise, it also usually introduces an annoying "musical noise". We have devised a method that eliminates this "musical noise" while further reducing the background noise. The method consists in subtracting an overestimate of the noise power spectrum, and preventing the resultant spectral components from going below a preset minimum level (spectral floor). The method can automatically adapt to a wide range of signal-to-noise ratios, as long as a reasonable estimate of the noise spectrum can be obtained. Extensive listening tests were performed to determine the quality and intelligibility of speech enhanced by our method. Listeners unanimously preferred the quality of the processed speech. Also, for an input signal-to-noise ratio of 5 dB, there was no loss of intelligibility associated with the enhancement technique.

Oscillating wave partial discharge detection technology has been widely used because of its powerful applicability. However, the detection process will produce short duration, burst, non-stationary impulse noise, whose characteristics are similar to partial discharge signals, and it is difficult to denoise by traditional methods such as filtering. In accordance with the above problems, this paper presents a method of recognition and noise reduction of oscillating partial discharge signal based on support vector machine. A classifier of noise pulse and partial discharge signal is established to eliminate impulse noise in partial discharge signal. Aiming at the difference of time-frequency characteristics between oscillating partial discharge signal and impulse noise, support vector machine classifier is used to identify partial discharge signal and impulse noise. It can overcome the defect that the traditional filtering method is difficult to suppress impulse noise. The experimental results show that, compared with the traditional empirical threshold method, the accuracy of partial discharge defect location is improved by more than 7.5%, which can effectively improve the accuracy of partial discharge location detection.

Acoustic emission (AE) detection, as a non-electrical detection method, is very suitable for effective fault detection of power equipment with a strong electromagnetic field. However, the AE signal collected at industrial sites often contains a lot of interference noise, affecting the analysis and prediction of faults. In this study, a wavelet denoising method based on a new threshold function is proposed, to achieve a noise reduction in the low signal-to-noise ratio (SNR) signals. Simulation experiment results show that the proposed threshold function not only overcomes the shortcomings of the discontinuous hard threshold function, but also solves the constant deviation of the soft threshold function. What’s more, the proposed function achieves a good adaptability. When SNR = 10 dB: The SNR of the new threshold is 20.6622, and the RMSE is 0.0026. The SNR of the hard threshold is 20.2246 and the RMSE is 0.0027, compared with the traditional hard threshold method, the SNR of the new threshold is increased by 2.16% and the root mean square error (RMSE) is reduced by 3.7%; the SNR of the soft threshold is 15.5656, and the RMSE is 0.0047, compared with the traditional soft threshold method, the new threshold has a 32.74% increase in SNR and a 40.43% reduction in RMSE. When SNR = -10 dB: The SNR of the new threshold is 4.2602, and the RMSE is 0.0172. The SNR of the hard threshold is 3.8558 and the RMSE is 0.0182, compared with the traditional hard threshold method, the SNR of the new threshold is increased by 10.49% and the RMSE is reduced by 5.49%; the SNR of the soft threshold is 2.1625, and the RMSE is 0.0212, compared with the soft threshold method, SNR is improved by 97% and RMSE is reduced by 18.87%. Performance analyses have proved that the improved wavelet denoising method can obtain a good noise reduction effect. It is very helpful for AE signal analysis with the generally low SNR, which can improve the accuracy of failure identification in subsequent acts.

Traditional speech enhancement algorithms have been widely studied due to simplicity and convenience, especially spectral subtraction algorithm. However, traditional speech enhancement algorithms always utilize the available noisy speech phase spectrum as the optimal estimation of clean speech phase spectrum, which is inaccurate at low signal-noise ratio (SNR). Aiming at improving performance of typical spectral subtraction speech enhancement algorithm under the circumstances of low SNR, a multiband spectral subtraction speech enhancement algorithm with phase spectrum compensation is proposed in this paper. The proposed method compensates noisy speech phase spectrum based on the SNR and estimation of noise obtained in the current frame and reduces musical noise generated by spectral subtraction. Simulation results show that the proposed algorithm has a better performance than traditional algorithms, especially at low SNR.

Due to traditional classification methods can’t meet the requirements of a great variety of cardiovascular diseases in clinical environment, CNN recognition method based on wavelet denoising is proposed to apply in automatic analysis of multi-category heart sounds. Firstly, original heart sound signal is denoising by wavelet threshold and the Mel-Frequency Spectrum Coefficients are extracted as the features of heart sound. Then models of CNN and Long Short-Term Memory (LSTM) networks are constructed and the related parameters are adjusted. Finally, the trained models are used to classify the heart sound signals. The experimental results show all performance indicators including sensitivity of proposed algorithm outperform LSTM and the traditional methods, achieves a better recognition results at the same time.

Speech is easily interfered by external environment in reality, which results in the loss of important features. Deep learning has become a popular speech enhancement method because of its superior potential in solving nonlinear mapping problems for complex features. However, the deficiency of traditional deep learning methods is the weak learning capability of important information from previous time steps and long-term event dependencies between the time-series data. To overcome this problem, we propose a novel speech enhancement method based on the fused features of deep neural networks (DNNs) and gated recurrent unit (GRU). The proposed method uses GRU to reduce the number of parameters of DNNs and acquire the context information of the speech, which improves the enhanced speech quality and intelligibility. Firstly, DNN with multiple hidden layers is used to learn the mapping relationship between the logarithmic power spectrum (LPS) features of noisy speech and clean speech. Secondly, the LPS feature of the deep neural network is fused with the noisy speech as the input of GRU network to compensate the missing context information. Finally, GRU network is performed to learn the mapping relationship between LPS features and log power spectrum features of clean speech spectrum. The proposed model is experimentally compared with traditional speech enhancement models, including DNN, CNN, LSTM and GRU. Experimental results demonstrate that the PESQ, SSNR and STOI of the proposed algorithm are improved by 30.72%, 39.84% and 5.53%, respectively, compared with the noise signal under the condition of matched noise. Under the condition of unmatched noise, the PESQ and STOI of the algorithm are improved by 23.8% and 37.36%, respectively. The advantage of the proposed method is that it uses the key information of features to suppress noise in both matched and unmatched noise cases and the proposed method outperforms other common methods in speech enhancement.

A speech enhancement system for noisy speech signals under different environments, namely airport, babble, car, exhibition, restaurant, station, street and train, is designed and implemented in this paper. Spectral subtraction is the most popular method for speech enhancement which suffers from perceptually annoying musical noise. Musical noise is reduced here by using low variance spectrum estimators and enhancement is achieved by wavelet thresholding the multitaper spectrum of the speech. The main feature of the proposed system is a novel switching mechanism to an optimal wavelet filter bank or wavelet packet filter bank matching to the critical bands of the human ear, based on the type of noise present in the input speech. The system is evaluated using noisy speech signals under different environment noises at different signal to noise ratios. The subjective and objective test results show that the proposed method improves the quality and intelligibility of speech signals.

Ball mill safety and stability is of great significance in industrial production. As a typical case of the signals with multiple mixed similar vibration forms and meshing noise, ball mill jars (BMJ), is selected to diagnosis and analysis in detail. The traditional method of fault diagnosis of gear mesh noise has high requirements on the machine operating environment, and is insensitive to the characteristics of the acoustic signal that is not tightly fitted. A new method, the wavelet denoising and combined with power spectrum, is proposed for the signal that BMJ and base are not tightly fit to produce a piecewise polynomial structure acoustic signal. The acoustic vibration signal is denoised by wavelet, and then the filtered signal is processed by the AR model power spectrum. By analyzing the characteristic power spectrum image, the unfit state can be diagnosed with high accuracy. In order to verify the effectiveness of the method, it is compared with the processed power spectrum image after the noise signal is processed. Experimental results show that the method has better diagnostic performance. The method is an automatic fault diagnosis and early warning system with high precision and robustness in noisy environments.

Wavelet denoising is commonly used for speech enhancement because of the simplicity of its implementation. However, the conventional methods generate the presence of musical residual noise while thresholding the background noise. The unvoiced components of speech are often eliminated from this method. In this paper, a novel algorithm of wavelet coefficient threshold (WCT) based on time-frequency adaptation is proposed. In addition, an unvoiced speech enhancement algorithm is also integrated into the system to improve the intelligibility of speech. The wavelet coefficient threshold (WCT) of each subband is first temporally adjusted according to the value of a posterior signal-to-noise ratio (SNR). To prevent the degradation of unvoiced sounds during noise, the algorithm utilizes a simple speech/noise detector (SND) and further divides speech signal into unvoiced and voiced sounds. Then, we apply appropriate wavelet thresholding according to voiced/unvoiced (V/U) decision. Based on the masking properties of human auditory system, a perceptual gain factor is adopted into wavelet thresholding for suppressing musical residual noise. Simulation results show that the proposed method is capable of reducing noise with little speech degradation and the overall performance is superior to several competitive methods.

Radio interferometers are used to construct high resolution images of the sky at radio frequencies and are the key instruments for accessing the statistical properties of the evolution of neutral hydrogen over cosmic time. Here we use simulated observations of the model sky to assess the efficacy of different estimators of the large-scale structure and power spectrum of the sky brightness distribution. We find that while the large-scale distribution can be reasonably estimated using the reconstructed image from interferometric data, estimates of the power spectrum of the intensity fluctuations calculated from the image are generally biased. This bias is found to be more pronounced for diffuse emission. The visibility based power spectrum estimator, however, gives an unbiased estimate of the true power spectrum. This work demonstrates that for an observation with diffuse emission the reconstructed image can be used to estimate the large-scale distribution of the intensity, while to estimate the power spectrum, visibility based methods should be preferred. With the upcoming experiments aimed at measuring the evolution of the power spectrum of the neutral hydrogen distribution, this is a very important result.