Nonnegative matrix factorization (NMF) is one of the most popular machine learning tools for speech enhancement (SE). However, there are two problems reducing the performance of the traditional NMFbased SE algorithms. One is related to the overlap-and-add operation used in the short time Fourier transform (STFT) based signal reconstruction, and the other is the Euclidean distance used commonly as an objective function; these methods can cause distortion in the SE process. In order to get over these shortcomings, we propose a novel SE joint framework which combines the discrete wavelet packet transform (DWPT) and the Itakura-Saito nonnegative matrix factorisation (ISNMF). In this approach, the speech signal was first split into a series of subband signals using the DWPT. Then, the ISNMF was used to enhance the speech for each subband signal. Finally, the inverse DWPT (IDWT) was utilised to reconstruct these enhanced speech subband signals. The experimental results show that the proposed joint framework effectively enhances the performance of speech enhancement and performs better in the unseen noise case compared to the traditional NMF methods.
Biometrics provide an alternative to passwords and pins for authentication. The emergence of machine learning algorithms provides an easy and economical solution to authentication problems. The phases of speaker verification protocol are training, enrollment of speakers and evaluation of unknown voice. In this paper, we addressed text independent speaker verification using Siamese convolutional network. Siamese networks are twin networks with shared weights. Feature space can be learnt easily by training these networks even if similar observations are placed in proximity. Extracted features from Siamese then can be classified using difference or correlation measures. We have implemented a customized scoring scheme that utilizes Siamese’ capability of applying distance measures with the convolutional learning. Experiments made on cross language audios of multi-lingual speakers confirm the capability of our architecture to handle gender, age and language independent speaker verification. Moreover, our designed Siamese network, SpeakerNet, provided better results than the existing speaker verification approaches by decreasing the equal error rate to 0.02.
For many years, a digital waveguide model is being used for sound propagation in the modeling of the vocal tract with the structured and uniform mesh of scattering junctions connected by same delay lines. There are many varieties in the formation and layouts of the mesh grid called topologies. Current novel work has been dedicated to the mesh of two-dimensional digital waveguide models of sound propagation in the vocal tract with the structured and non-uniform rectilinear grid in orientation. In this work, there are two types of delay lines: one is called a smaller-delay line and other is called a larger-delay line. The larger-delay lines are the double of the smaller delay lines. The scheme of using the combination of both smaller- and larger-delay lines generates the non-uniform rectilinear two-dimensional waveguide mesh. The advantage of this approach is the ability to get a transfer function without fractional delay. This eliminates the need to get interpolation for the approximation of fractional delay and give efficient simulation for sound wave propagation in the two-dimensional waveguide modeling of the vocal tract. The simulation has been performed by considering the vowels /ɔ/, /a/, /i/ and /u/ in this work. By keeping the same sampling frequency, the standard two-dimensional waveguide model with uniform mesh is considered as our benchmark model. The results and efficiency of the proposed model have compared with our benchmark model.
Noise exposure is one of the most important physical agents in the workplace which can induce job stress in several ways. The aim of this study was to model the interactions between independent and mediating variables and job stress using structural equation modeling. In this study, Weinstein’s noise sensitivity scale, noise annoyance questionnaire, Health and Safety Executive (HSE) job stress questionnaire and job satisfaction scale were used. To assess worker’s noise exposure, the 8-hours equivalent continuous A-weighted sound pressure level (LAeq;8 h), was measured based on ISO 9612 (2009). To achieve the aims of study, the structural equation model was run using R software 3.4.1 and Cytoscape software 3.6.0. Based on the results, while there was a direct positive correlation of noise exposure on total job stress, there were also indirect positive effects through job satisfaction and noise sensitivity as mediator variables. Using hearing protective devices negatively affected total job stress through a direct pathway and an indirect pathway when job satisfaction was a mediator variable. Regarding the total effect of noise exposure and using hearing protection devices on job stress subscales, it can be concluded that noise exposure and using hearing protection devices had greatest effect on colleagues support and demand, respectively. It can be concluded that noise exposure and lack of hearing protective devices have a significant positive effect on job stress among workers of a textile industry. In addition to the direct effect, this factor can induce job stress through noise sensitivity, job satisfaction and noise annoyance. Therefore, measures which can decrease any of the mentioned factors, also can alleviate job stress.
The minimum size of the bootstrap algorithm input parameters have been determined for estimation of long-term indicators of road traffic noise. Two independent simulation experiments have been performed for that purpose. The first experiment served to determine the impact of original random sample size, and the second to determine the impact of number of the bootstrap replications on the accuracy and uncertainty of estimation of long-term noise indicators. The inference has been carried out based on results of non-parametric statistical test at significance level α = 0.05. The simulation experiments have shown that estimation of long-term noise indicators with uncertainty below ±1 dB(A) requires all-day noise measurements during three randomly selected days during the year in a dense urban development. The maximum size of original random sample should not exceed n = 50 elements. The minimum number of bootstrap replications necessary for estimation should be B = 5000. The data used to the simulation experiments and carry out the analysis were results of continuous monitoring of road traffic noise recorded in 2009 in one of the main arteries of Krakow in Poland.
The noise perceiving issue is very subjective and depends on several factors, such as: the living environment in which each person has grown and developed, the education they have received, the culture in which their life principles have formed and, last but not least, the social and financial status. Therefore, in order to establish effective actions in multiple directions when it comes to any urban noise analysis, it is very important to know the perception and the subjective reactions of the individuals involved. The paper respects this idea, presenting the results of a sociological study on urban noise, applied in the city of Cluj-Napoca, Romania. The intention was to capture the reactions of the permanent residents of the city, but also of the people in transit, as well as to analyse the changes that occurred as result of the implementation of the Environmental Noise Directive (European Commission). The study shows that 75.2% of the respondents consider that the noise in the city has increased in the last ten years and 58% of them have rated the noise as level 4 or 5 on a five point scale. Information related to noise maps and actions taken to reduce community noise has no sufficient dissemination. There is also a medium to low reaction of the population in correlation to the declared noise annoyance.
One of the main issues of design process of HVAC systems and ventilation ducts in particular is correct modelling of coupling of the flow field and acoustic field of the air flowing in such systems. Such a coupling can be modelled in many ways, one of them is using linearised Euler equations (LEE). In this paper, the method of solving these equations using finite element method and open source tools is decribed. Equations were transformed into functional and solved using Python language and FEniCS software. The non-reflective boundary condition called buffer layer was also implemented into equations, which allowed modelling of unbounded domains. The issue, influence of flow on wave propagation, could be adressed using LEE equations, as they take non-uniform mean flow into account. The developed tool was verified and results of simulations were compared with analytical solutions, both in one- and two-dimensional cases. The obtained numerical results are very consistent with analytical ones. Furthermore, this paper describes the use of the developed tool for analysing a more complex model. Acoustic wave propagation for the backward-facing step in the presence of flow calculated using Navier-Stokes equations was studied.
Two optimization aspects of the meshless method (MLM) based on nonsingular radial basis functions (RBFs) are considered in an acoustic indoor problem. The former is based on the minimization of the mean value of the relative error of the solution in the domain. The letter is based on the minimization of the relative error of the solution at the selected points in the domain. In both cases the optimization leads to the finding relations between physical parameters and the approximate solution parameters. The room acoustic field with uniform, impedance walls is considered.
As results, the most effective Hardy’s Radial Basis Function (H-RBF) is pointed out and the number of elements in the series solution as a function of frequency is indicated. Next, for H-RBF and fixed n, distributions of appropriate acoustic fields in the domain are compared. It is shown that both aspects of optimization improve the description of the acoustic field in the domain in a strictly defined sense.
In the present work, the radiation of sound waves from a coaxial duct is considered. This coaxial duct has an inner wall which is infinite and has piecewise acoustically absorbent material, while the outer wall is semi-infinite and rigid. The analytical solution of the problem is found by means of the Wiener-Hopf technique. Applying the Fourier transformation to the boundary value problem, the explicit expression for the scattered field is obtained. In the end, some numerical results are displayed for different parameters and compared to rigid case.
The evaluation of complex radiation impedance for a square piston source on an infinite circularcylindrical baffle is associated to the Greenspon-Sherman formulation for which novel evaluation methods are proposed. Unlike existing methods results are produced in a very wide range of frequencies and source semi-angles with controllable precision. For this reason closed-form expressions are used to describe the truncation errors of all integrals and infinite sums involved. Impedance values of increased accuracy are also provided in tabulated form for engineering use and a new radiation mass-load model is derived for low-frequencies.
The paper presents a measuring system based on two resonators with a SAWacoustic surface wave. One of the resonators contains a sensor structure consisting of a Nafion layer with a PANI polyaniline nanolayer deposited on it. The sensor structure was tested for carbon monoxide, with a very low concentration (5, 10, 15, 20 ppm) in the atmosphere of synthetic air. The structure sensitivity was tested for two different PANI thicknesses: (100 and 180 nm). The tests were carried out for two different temperatures: 308 K and 315 K. The investigations shows that the measuring system used with the acoustic surface wave together with the proposed sensing layers is sensitive to the presence of low concentration carbon monoxide molecules in the atmosphere of synthetic air.
Based on the electromechanical equivalent circuit theory, equations related to the resonance frequency and the magnifying coefficient of a quarter-wave vibrator and a quarter-wave taper transition horn were deduced, respectively. A series of 3D models of ultrasonic composite transducers with various conical section length was also established. To reveal the influences of the conical section length and the prestressed bolt on the dynamic characteristics (resonance frequency, amplitude, displacement node, and the maximum equivalent stress) of the models and the design accuracy, finite element (FE) analyses were carried out. The results show that the addition of prestressed bolt increases the resonance frequency and causes the displacement node on the center axis to move towards the small cylindrical section. As the conical section length rises, the increment of resonance frequency reduces and tends to a stable value of 360 Hz while the displacement of the node on the center axis becomes lager and gradually approaches 1.5 mm. Furthermore, the amplitude of the output terminal is stable at 16.18 μm under 220 V peak-topeak (77.8 VRMS) sinusoidal potential excitation. After that, a prototype was fabricated and validated experiments were conducted. The experimental results are consistent with that of theory and simulations. It provides theoretical basis for the design and optimization of small-size, large-amplitude, and high-power composite transducers.
One major problem in the design of ultrasonic transducers results from a huge impedance mismatch between piezoelectric ceramics and the loading medium (e.g. gaseous, liquid, and biological media). Solving this problem requires the use of a matching layer (or layers). Optimal selection of materials functioning as matching layers for piezoelectric transducers used in transmitting and receiving ultrasound waves strictly depends on the type of the medium receiving the ultrasound energy. Several methods allow optimal selection of materials used as matching layers. When using a single matching layer, its impedance can be calculated on the basis of the Chebyshev, DeSilets or Souquet criteria. In the general case, the typically applied methods use an analogy to a transmission line in order to calculate the transmission coefficient T. This paper presents an extension of transmission coefficient calculations with additional regard to the attenuation coefficients of particular layers. The transmission coefficient T is optimised on the basis of a genetic algorithm method. The obtained results indicate a significant divergence between the classical calculation methods and the genetic algorithm method.
DIFAR type underwater passive systems are one of the more commonly used tools for detecting submarines. At the design stage, which usually uses computer simulations, it is necessary to generate acoustic noise of the sea. It has been shown that correlating noise significantly reduces these errors compared to the assumption that noise is uncorrelated. In addition, bearing errors have been shown to be the same in systems with a commonly used antenna containing five hydrophones, as in a system without a central hydrophone, which may be useful in some DIFAR system design solutions.
Marine mammal identification and classification for passive acoustic monitoring remain a challenging task. Mainly the interspecific and intraspecific variations in calls within species and among different individuals of single species make it more challenging. Varieties of species along with geographical diversity induce more complications towards an accurate analysis of marine mammal classification using acoustic signatures. Prior methods for classification focused on spectral features which result in increasing bias for contour base classifiers in automatic detection algorithms. In this study, acoustic marine mammal classification is performed through the fusion of 1D Local Binary Pattern (1D-LBP) and Mel Frequency Cepstral Coefficient (MFCC) based features. Multi-class Support Vector Machines (SVM) classifier is employed to identify different classes of mammal sounds. Classification of six species named Tursiops truncatus, Delphinus delphis, Peponocephala electra, Grampus griseus, Stenella longirostris, and Stenella attenuate are targeted in this research. The proposed model achieved 90.4% accuracy on 70–30% training testing and 89.6% on 5-fold cross-validation experiments.
This paper presents comparative analysis of various acoustic signals expected during partial discharge (PD) measurements in operating power transformer. Main purpose of the paper is to yield relevant and reliable method to distinguish between various acoustic emission (AE) signals emitted by PD and other sources, with particular consideration of real-life results rather than laboratory simulations. Therefore, selected examples of real-life AE signals registered in seven different power transformers, under normal operation conditions, within few years are showed and analyzed. Five scenarios are investigated, which represent five types of AE sources: PD generated by artificial sources, and next four real-life sources (including PD in working transformer, oil flow, oil pumps and core). Several different signal processing methods are applied and compared in order to identify the PD signals. As a result, an energy patterns analysis based on the wavelet decomposition is found as the most reliable tool for identification of PD signals. The presented results may significantly support the process of interpretation of the PD measurement results, and may be used by field engineers as well as other researchers involved in PD analysis using AE method. Finally, observed properties also provide a solid basis for establishing or improving complete classification method based on the artificial intelligence algorithms.
The article discusses the issues of accounting the direction pattern of parametric antenna array the propagation of sound over the Earth’s surface. As a radiator, a parametric antenna array is used. A description is given of measuring equipment and experimental research methods. The Delaney-Bezley model was used as a model of the Earth’s surface impedance. The research results showed the importance of accounting the direction pattern of parametric antenna array in predicting the sound pressure level of a propagating acoustic signal over the Earth’s surface. On the example of a difference signal with a frequency of 2 kHz, the calculation of the sound pressure level on a 100-meter path with the influence of the Earth’s surface is shown. The results obtained showed a good agreement between the theoretical calculation and experimental data.
Radial basis function neural networks (RBF NNs) are one of the most useful tools in the classification of the sonar targets. Despite many abilities of RBF NNs, low accuracy in classification, entrapment in local minima, and slow convergence rate are disadvantages of these networks. In order to overcome these issues, the sine-cosine algorithm (SCA) has been used to train RBF NNs in this work. To evaluate the designed classifier, two benchmark underwater sonar classification problems were used. Also, an experimental underwater target classification was developed to practically evaluate the merits of the RBFbased classifier in dealing with high-dimensional real world problems. In order to have a comprehensive evaluation, the classifier is compared with the gradient descent (GD), gravitational search algorithm (GSA), genetic algorithm (GA), and Kalman filter (KF) algorithms in terms of entrapment in local minima, the accuracy of the classification, and the convergence rate. The results show that the proposed classifier provides a better performance than other compared classifiers as it classifies the sonar datasets 2.72% better than the best benchmark classifier, on average.
Professor Andrzej Orłowski, a long-time employee of National Marine Fisheries Research Institute, passed away on October 27, 2020. He was an outstanding scientist in the field of hydroacoustics. Inventor of the method of use of multiple echo measurements to assess the type of the seabed. To this day, this method is called the Orłowski Method. Professor Orłowski was a member of the Physics Section of the SCOR, the NMFRI Scientific Council, the Polish Acoustical Society, ICES Fishery Acoustic Science and Technology Group, ICES Fish Technology and Fish Behavior Working Group, ICES SG Acoustic Seabed Classification and ICES SG Fish Avoidance of Research Vessels.
To subscribe to the magazine enter the email address:
*Fields marked with an asterisk are mandatory to be filled in and checked. To Subscribe to the journal you must agree to the processing of personal data.