Multi-band Spectral Subtraction Research Articles

Comparative Study on Noisy Speech Preprocessing Algorithms

Speech Enhancement and noise reduction have wide applications in speech processing. They are often employed as pre-processing stage in various applications. The work to be presented in this paper is denoising a single-channel speech signal at the presence of a highly non-stationary background noise in order to improve the perceptible quality and intelligibility of the speech. Real world noise is mostly highly non-stationary and does not affect the speech signal uniformly over the spectrum. This paper investigates various Discrete Fourier Transform-based algorithms as single-channel pre-processing techniques consisting of: Spectral Subtraction using over-subtraction and spectral floor, Multi-Band Spectral Subtraction (MBSS), Wiener Filter, MMSE of Short-Time Spectral Amplitude (MMSE-STSA) estimator with, and without using SPU modifier, MMSE Log-Spectral Amplitude Estimator with, and without using SPU modifier, Optimally-Modified Log-Spectral Amplitude estimator (OM-LSA). The processed speeches from these algorithms are compared at the same set of conditions using visual examinations of signals in the time domain and the spectrograms, and also the objective and subjective tests for quality and perceptual evaluation. All the implemented algorithms provide considerable, different degrees of flexibility and control on noise elimination levels that reduces artifacts in the enhanced speech, resulting in the improved quality, and intelligibility.

Mitigating Information Interruptions by COVID-19 Face Masks: A Three-Stage Speech Enhancement Scheme

The coronavirus disease 2019 (COVID-19) preventive measures have resulted in significant lifestyle changes. One of the COVID-19 new normal is the usage of face masks for protection against airborne aerosol which creates distractions and interruptions in voice communication. It has a different influence on speech than the standard concept of noise affecting speech communication. Furthermore, it has varied effects on speech in different frequency bands. To provide a solution to this problem, a three-stage adaptive speech enhancement (SE) scheme is developed in this article. In the first stage, the tunable <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$Q$</tex-math> </inline-formula> -factor wavelet transform (TQWT) features are extracted by properly setting the quality factor values and the number of levels from the input speech signal. In the second stage, the adjustable parameters of the preemphasis filter and modified multiband spectral subtraction (MBSS) are determined using bio-inspired techniques for different masking and signal-to-noise ratio (SNR) conditions. In the third stage, the weights, center values, standard deviation of the Gaussian radial basis functions, and input patterns of the radial basis function neural networks (RBFNNs) are updated to predict the optimized parameters from the input TQWT-based cepstral features (TQCFs). In the end, the performance of the proposed algorithm is compared with the standard SE algorithms using two speech datasets.

Classification of pathological disorders using optimization enabled deep neuro fuzzy network

The discovery of voice pathology is an emerging domain in biomedical engineering. The assessment helps to offer information for effective diagnosis. An optimization-enabled deep model for pathological disorder classification is devised in this paper. The speech signal is employed as the input of the pre-processor which subsequently helps to eliminate noise considering the Hanning window. The initial improvement is done by exploiting multi-band spectral subtraction for generating frames. The enhancement of speech is performed using Deep Residual Network (DRN) classifier. The training of DRN is performed by exploiting adopted Feedback Artificial Tree Firefly algorithm (FATFA) that is attained by combining Feedback Artificial Tree Algorithm (FAT), and Firefly algorithm (FA). The features, like Amplitude modulation spectrogram (AMS), and frequency-domain features namely tonal power ratio, pitch chroma, spectral spread, spectral flux, as well as spectral autocorrelation is mined from the improved speech. The classification is done with Deep Neuro-Fuzzy Network (DNFN), which is tuned with the devised Verse FATFA. The proposed Verse FATFA is conceived by merging FATFA and Multiverse Optimizer (MVO). Therefore, the devised Verse FATFA-based DNFN categories voice disorders into Laryngitis, Dysphonie, spasmodischedysphonie, Leukoplakie, Cyste, Dish-Syndrom, and Monochorditis. The proposed Verse FATFA-based DNFN provides enhanced performance with maximum values of 94.93% accuracy, 95.94% sensitivity with 91.93% specificity.

A Two-Stage Deep Neuroevolutionary Technique for Self-Adaptive Speech Enhancement

This paper presents a novel self-adaptive approach for speech enhancement in the context of highly nonstationary noise. A two-stage deep neuroevolutionary technique for speech enhancement is proposed. The first stage is composed of a deep neural network (DNN) method for speech enhancement. Two DNN methods were tested at this stage, namely, both a deep complex convolution recurrent network (DCCRN) and a residual long short-term memory neural network (ResLSTM). The ResLSTM method was combined with a minimum mean-square error method to perform a preliminary enhancement. The ResLSTM network is used as an <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">a priori</i> signal-to-noise ratio (SNR) estimator. The second stage implements a self-adaptive multiband spectral subtraction enhancement method using tuning optimization based on a genetic algorithm. The proposed two-stage technique is evaluated using objective measures of speech quality and intelligibility. The experiments are carried out using the NOIZEUS noisy speech corpus using conditions of real-world stationary, colored, and nonstationary noise sources at multiple SNR levels. These experiments demonstrate the advantage of building a cooperative approach using evolutionary and deep learning-based techniques that are capable of achieving robust speech enhancement in adverse conditions. Indeed, the experimental tests show that the proposed two-stage technique outperformed a baseline implementation using a state-of-the-art deep learning approach by an average 13% and 6% improvement for six noise conditions at a −5 dB and a 0 dB input SNR, respectively.

Acoustical and Perceptual Analysis of Noise Reduction Strategies in Individuals With Auditory Neuropathy Spectrum Disorders.

Purpose The conventional amplification devices render minimal or no benefit at abating the speech perception problems of individuals with auditory neuropathy spectrum disorder (ANSD). This study was undertaken to evaluate the effect of noise reduction strategies (multiband spectral subtraction, Wiener-as, Karhunen-Loeve transform [Subspace], and ideal binary mask [IdBM] algorithm) on speech using speech perception measures and acoustic measure among individuals with ANSD. Method Two groups of participants (age: 17-43 years) were recruited in the study. Group I comprised 12 individuals with a confirmed diagnosis of ANSD and not exceeding moderate degree of hearing loss and Group II of 10 individuals with normal hearing in both ears. The signal-to-noise required for 50% speech recognition (SNR-50) was measured for the participants in five conditions, that is, unprocessed speech and speech processed with four noise reduction strategies. Additionally, an acoustic objective measure Extended Short-Time Objective Intelligibility algorithm was employed to estimate the intelligibility index across the conditions. Results Significant difference was found across conditions in both the groups. Pairwise comparison revealed significantly better speech perception on SNR-50 measure with IdBM strategy, for both the groups. No significant difference in SNR-50 was observed with other noise reduction strategies. IdBM condition also gave the highest intelligibility index (d) values using Extended Short-Time Objective Intelligibility algorithm. This finding needs to be verified on a larger group of individuals with ANSD. Conclusions IdBM noise reduction strategy rendered significantly lower SNR-50 compared to other noise reduction strategies for individuals with ANSD in this study. This provides clinical evidence for the same and also recommends trying on a larger group of participants before its implementation in hearing devices. Apart from this, the current strategies used in hearing aids provide no improvement in speech identification in noise for this population. Hence, though the present hearing aids may show benefit in quiet condition, chances of its rejection are high in noisy backgrounds.

Speech Intelligibility Based Enhancement System Using Modified Deep Neural Network and Adaptive Multi-band Spectral Subtraction

In contrast to the adverse environments, performances of existing speech enhancement algorithms do not always produce satisfactory results. In the case of worst signal to noise ratio, the processing is complicated and it may introduce signal distortions and degradation of intelligibility. To overcome the complexity of the existing speech enhancement algorithms, a hybrid concept for enhancing the speech quality and intelligibility is proposed in this research. The primary objectives of the research work is to increase the intelligibility of the speech enhancement system that has been trained for a particular speech signal using modified deep neural network (DNN) and adaptive multi-band spectral subtraction (AdMBSS). In this work, AdMBSS is used for enhancing the intelligibility of the speech signal using the additional phase information calculation, and finally, hybrid DNN and Nelder Mead optimization is utilized to improve the signal quality. Experimental results explain that the proposed framework achieves improved performance in signal to noise ratio, perceptual evaluation of signal quality and minimum mean square error. Finally, performances are taken for the more noises like bus noise, train noise, babble noise, airport noise, station noise and exhibition noise.

Insight on the utilization of the noise estimation techniques in phase aware modified multiband spectral subtraction

ABSTRACTThe standard technique of all Spectral Subtraction algorithms that are used for Speech Enhancement (SE) is to find out the clean speech by subtracting the noise estimate from the noisy speech spectrum. So the essential criterion is the accuracy of noise estimation (NE) and it affects the potency of the SE. The traditional NE algorithms use voice activity detection (VAD) for noise estimation, which is troublesome at the cases of low SNR condition and when the noise characteristics are not stationary. The effect of noise is also different in different frequency bands of the noisy speech spectrum. In this paper, several NE techniques are implemented along with a phase aware and frequency dependent modified multi-band spectral subtraction. To give additional intelligibility to SE, the phase of the enhanced signal is modified using the phase spectrum compensation function. Finally, the mentioned methods are simulated and compared using several objective evaluation measures with the help of standard speech database.

Multi Band Spectral Subtraction Combined with Adaptive Noise Estimation

Speech enhancement is a prerequisite for many speech communication applications like speech coders, hearing aids, voice recognition systems, Man –Machine communication, Video conferencing etc . For all these practical systems input is a noisy speech , so it requires a pre-processing operation for further processing. This paper gives a method to improve the speech quality in presence of background noise, based on Multi Band Spectral Subtraction combined with adaptive noise estimation. Further the performance analysis of this method is done using objective quality measures like Segmental SNR(segSNR),Weighted SlopeSpectral Distance (WSSD),Log Likelihood Ratio(LLR),Perceptual Evaluation of Speech Quality (PESQ) and Composite Measures under highly non-stationary noise. Finally the results were compared against Spectral Subtraction(SS), Multi Band Spectral Subtraction(MBSS) methods and the results shows that the proposed method shows consistence performance against all the cases considered.

Multi Band Spectral Subtraction for Speech Enhancement with Different Frequency Spacing Methods and their Effect on Objective Quality Measures

Multi Band Spectral Subtraction for Speech Enhancement with Different Frequency Spacing Methods and their Effect on Objective Quality Measures

Feature Enhancement Method for Drilling Vibration Signal by Using Wavelet Packet Multi-band Spectral Subtraction

Feature Enhancement Method for Drilling Vibration Signal by Using Wavelet Packet Multi-band Spectral Subtraction

Speech intelligibility enhancement for Thai-speaking cochlear implant listeners

Cochlear implant (CI) listeners encounter difficulties in communicating with other persons in noisy listening environments. However, most CI research has been carried out using the English language. In this study, single-channel speech enhancement (SE) strategies as a pre-processing approach for the CI system were investigated in terms of Thai speech intelligibility improvement. Two SE algorithms, namely multi-band spectral subtraction (MBSS) and Weiner filter (WF) algorithms, were evaluated. Speech signals consisting of monosyllabic and bisyllabic Thai words were degraded by speech-shaped noise and babble noise at SNR levels of 0, 5, and 10 dB. Then the noisy words were enhanced using SE algorithms. The enhanced words were fed into the CI system to synthesize vocoded speech. The vocoded speech was presented to twenty normal-hearing listeners. The results indicated that speech intelligibility was marginally improved by the MBSS algorithm and significantly improved by the WF algorithm in some conditions. The enhanced bisyllabic words showed a noticeably higher intelligibility improvement than the enhanced monosyllabic words in all conditions, particularly in speech-shaped noise. Such outcomes may be beneficial to Thai-speaking CI listeners.

FPGA based dual microphone speech enhancement

This paper proposes an efficient reconfigurable hardware design of dual microphone speech enhancement technique using sound source localization and multi band spectral subtraction methods with elimination of background noise. Firstly, we have used a time delay of arrival algorithm using phase transform (PHAT) to achieve the time difference between the microphone signals. PHAT based filter can reach high SNR gains, which makes it very suitable for localizing the sound source in a microphone array system. After adjustment of the delay between the signals, multi band spectral subtraction technique enhances the signal from the background noise environment in each of the frequency bands. Our design has been implemented in Spartan6 Lx45 FPGA and we have presented the implementation results in terms of subjective and objective evaluations. We have also compared the angular separation between the microphones to the resultant angle between the microphones and observed that proposed design provides very sharp accuracy in every corresponding angle. The objective and subjective evaluation established that our design provides better throughput compared to the existing state of the art research works. The evaluation infers that our proposed hardware induce feasibility for hand-held devices in background noisy environment.

An Improved Multi-band Spectral Subtraction using Mel-scale

In order to improve the intelligibility and quality of speech signal under different noise environments, an improved multi-band spectral subtraction algorithm using Mel-scale is proposed in this paper. In this algorithm, the whole speech spectrum is divided into N subbands without overlapping each other using the Mel frequency domain, and uses spectral subtraction to obtain the estimated pure speech independently in each subband. By using various types of noise at various SNRs, the experimental data, including SNR, PESQ measure and non-professional hearing test, are obtained. The result show that the algorithm is superior to the multi-band spectral subtraction with uniform segmentation and the traditional spectral subtraction algorithm, without causing considerable signal distortion and remnant musical noise.

Pipelined Architecture of Multi-Band Spectral Subtraction Algorithm for Speech Enhancement

In this paper, a new pipelined architecture of the multi-band spectral subtraction algorithm has been proposed for real-time speech enhancement. The proposed hardware has been implemented on field programmable gate array (FPGA) device using Xilinx system generator (XSG), high-level programming tool, and Nexys-4 development board. The multi-band algorithm has been developed to reduce the additive colored noise that does not uniformly affect the entire frequency band of useful signal. All the algorithm steps have been successfully implemented on hardware. Pipelining has been employed on this hardware architecture to increase the data throughput. Speech enhancement performances obtained by the hardware architecture are compared to those obtained by MATLAB simulation using simulated and actual noises. The resource utilization, the maximum operating frequency, and power consumption are reported for a low-cost Artix-7 FPGA device.

Improved single channel phase‐aware speech enhancement technique for low signal‐to‐noise ratio signal

In the state-of-the-art single channel speech enhancement techniques, the short-time spectral amplitude is modified while the effect of the phase corruption due to the contamination of additive noise is neglected. This study introduces an improved speech enhancement algorithm based on a phase-aware multi-band spectral subtraction technique which estimates the spectral amplitude of the clean speech signal by considering the phase of the speech and noise signal components, and uses the estimated phase of the clean speech signal for signal reconstruction in the time domain. Experimental results show that the proposed algorithm yields better performance in terms of various objective and composite quality measures and other intelligibility assessment metrics while compared with other existing spectral subtraction methods. Using the composite objective measure quality evaluation technique, it is observed that the overall signal quality of the enhanced speech signal is improved on an average by 70% at 0 dB global input signal-to-noise ratio by using the proposed approach.

Speech Enhancement using Spectral Subtraction-type Algorithms: A Comparison and Simulation Study

Abstract The spectral subtraction is historically one of the first algorithms proposed for the enhancement of single channel speech. In this method, the noise spectrum is estimated during speech pauses, and is subtracted from the noisy speech spectrum to estimate the clean speech. This is also achieved by multiplying the noisy speech spectrum with a gain function and later combining it with the phase of the noisy speech. The drawback of this method is the presence of processing distortions, called remnant noise. A number of variations of the method have been developed over the past years to address the drawback. These variants form a family of spectral subtractive-type algorithms. The aim of this paper is to provide a comparison and simulation study of the different forms of subtraction-type algorithms viz. basic spectral subtraction, spectral over-subtraction, multi-band spectral subtraction, Wiener filtering, iterative spectral subtraction, and spectral subtraction based on perceptual properties. To test the performance of the subtractive-type algorithms, the objective measures (SNR and PESQ), spectrograms and informal listening tests are conducted for both stationary and non-stationary noises types at different SNRs levels. It is evident from the results that the modified forms of spectral subtraction method reduces remnant noise significantly and the enhanced speech contains minimal speech distortion.

Analysis and implementation of low‐power perceptual multiband noise reduction for the hearing aids application

Traditional noise reduction designs provide good performance but suffer from high complexity and long latency, which limits their application to hearing aids. Targeted for strict low-power and low-latency requirement of completely-in-the-canal type hearing aids, this study analyses and implements a previously proposed sample-based perceptual multiband spectral subtraction with a multiplication-based entropy voice activity detection. Simulation results reveal that the authors design can provide similar speech quality as others, but with lower computational complexity and simple control effort. The corresponding core-based architecture design further exploits processing characteristics of the proposed approach to reduce power consumption with a sign-magnitude and a preprocessed input data reuse scheme. Chip measurement shows that the design only consumes 83.7 µW at 0.6 V operation with 90 nm high threshold voltage (HVT) (high VT ) standard cell library.

Neuromorphic Pitch Based Noise Reduction for Monosyllable Hearing Aid System Application

This paper presents a low computational complexity hardware-oriented neuromorphic pitch based noise reduction (NR) algorithm and hardware implementation for monosyllable hearing aid system applications. The proposed NR design consists of a pitch-based voice activity detector (pitch-based VAD) for speech detection and a neuromorphic noise attenuator for speech enhancement. The pitch-based VAD is developed on ANSI S1.11 based filter bank architecture and employs the characteristics of monosyllable and nonlinear energy operator (NEO) to improve the accuracy of VAD. The neuromorphic noise attenuator reduces the background noise by using the characteristics of human hearing system and the clues of speech. Simulation results show that the proposed algorithm has better SNR and PESQ performance than other non-pitch based NR algorithms in non-stationary background noise environments. Compared with multiband (mband) spectral subtraction and minimum mean square error (mmse) algorithms, the computational complexity of the proposed algorithm can save 90% computational complexity. The hardware implementation consumes 47.74 μW at 0.5 V operation with 65 nm HVT standard cell library.

An MDCT-domain audio denoising method with a block switching scheme

In this paper, an audio denoising method is proposed for improving the quality of handheld audio recording devices. The proposed method reduces noise differently depending on the block size in the modified discrete cosine transform (MDCT) analysis of an audio coder. Specifically, denoising for a long block is performed by multi-band spectral subtraction (MBSS) with perceptually weighted scalefactor bands, while that for a short block is performed by subband power scaling to maintain coherence of power with the previously-denoised long block. In order to evaluate the performance of the proposed method, it is first embedded into MPEG-2 advanced audio coding (AAC) that is popularly used for audio recording devices. Then, its performance is compared with that of a conventional audio denoising method based on block thresholding in terms of cepstral distortion, subjective quality, and computational complexity. It is shown from performance comparison that the proposed method outperforms the block thresholding method in both objective and subjective measurements. Moreover, the complexity of the proposed method is sufficiently lowered to be implemented on most resource-constrained handheld audio recording devices, unlike the conventional method.

Mechanical noise suppression based on non-negative matrix factorization and multi-band spectral subtraction for digital cameras

This paper proposes a new non-stationary noise suppression method to reduce the mechanical noise generated when audio signals are recorded with a digital camera. The proposed method first utilizes a non-negative matrix factorization (NMF) technique to estimate the noise spectrum of the mechanical noise from the noisy audio spectrum. After that, the mechanical noise contaminated in the audio signal is suppressed by multi-band spectral subtraction. In particular, the NMF technique estimates the noise spectrum in a frame-wise manner in order for the proposed method to operate in real-time. The performance of the proposed mechanical noise suppression method is evaluated in terms of log-spectral distortion, cepstral distortion, subjective quality, and computational complexity. In addition, it is compared with the performance of conventional methods. It is shown from the evaluation that the proposed method provides lower log-spectral and cepstral distortions with better subjective preference than conventional methods. Moreover, the complexity of the proposed method is low enough to be implemented on a commercially available digital camera.