Sound Source Identification Research Articles

部分共有型Deep Neural Networkを用いた音源同定

部分共有型Deep Neural Networkを用いた音源同定

Iterative beamforming for identification of multiple broadband sound sources

The reconstruction of broadband sound sources is an important issue in industrial acoustics. In this paper, a model comprising multiple incoherent Gaussian random sources is considered. The aim is to estimate locations and powers of the sound sources using the pressures measured by an array of microphones. Each measured pressure is interpreted as a mixture of latent signals emitted by different sound sources. Then, an Iterative Beamforming (IB) method is developed to estimate the source parameters. This approach is based on the Expectation-Maximization (EM) algorithm, a well-known iterative procedure for solving maximum likelihood parameter estimation. More specifically, IB iteratively estimates the source contributions and performs beamforming on these estimates. In this work, experiments on real data illustrate the advantage of IB with respect to classical beamforming and Near-field Acoustical Holography (NAH). In particular, the proposed method is shown to work over a wider range of frequencies, to better estimate the source locations, and is able to quantify the powers of the sources. Furthermore, experiments illustrate that IB can not only localize the sources on a given surface, but also accurately estimate their 3D locations.

Perceptual Learning of Acoustic Noise Generates Memory-Evoked Potentials

Experience continuously imprints on the brain at allstages of life. The traces it leaves behind can produce perceptual learning [1], which drives adaptive behavior to previously encountered stimuli. Recently, it has been shown that even random noise, a type of sound devoid of acoustic structure, can trigger fast and robust perceptual learning after repeated exposure [2]. Here, by combining psychophysics, electroencephalography (EEG), and modeling, we show thatthe perceptual learning of noise is associated with evoked potentials, without any salient physical discontinuity or obvious acoustic landmark in the sound. Rather, the potentials appeared whenever amemory trace was observed behaviorally. Such memory-evoked potentials were characterized by early latencies and auditory topographies, consistent with a sensory origin. Furthermore, they were generated even on conditions of diverted attention. The EEG waveforms could be modeled as standard evoked responses to auditory events (N1-P2) [3], triggered by idiosyncratic perceptual features acquired through learning. Thus, we argue that the learning of noise is accompanied by the rapid formation of sharp neural selectivity to arbitrary and complex acoustic patterns, within sensory regions. Such a mechanism bridges the gap between the short-term and longer-term plasticity observed in the learning of noise [2, 4-6]. It could also be key to the processing of natural sounds within auditory cortices [7], suggesting that the neural code for sound source identification will be shaped by experience as well as by acoustics.

FINITE ELEMENT DETERMINATION OF THE HEAD-RELATED TRANSFER FUNCTION

The auditory peripheral system, which takes care of collection and enlargement in hearing processing, is an important part of the overall auditory system. This system is the key of sound source identification and location, and an important basis for a variety of audio equipment design and sound quality evaluation. In order to better recognize the acoustic characteristics of the auditory peripheral system, and effectively guide the audio equipment design as well as the sound quality evaluation, this study presents a numerical model based on the finite element/infinite element method (FEM/IFEM) to compute the near-field head-related transfer function (HRTF) on a reversed head model. A spherical sound source is positioned at different locations in the horizontal, median and lateral planes. The HRTFs and sound field distributions are computed for a frequency range up to 6 kHz. The present model is qualitatively validated against experimental data from MIT database. The results confirm the potentiality of the proposed approach which can efficiently detect the diffraction region, diffraction angle and tails in the sound field.

Sound source identification in a noisy environment based on inverse patch transfer functions with evanescent Green's functions

A modified inverse patch transfer function (iPTF) method is used to reconstruct the normal velocities of the target source in a noisy environment. The iPTF method simplifies the Helmholtz integral equation to one term by constructing a Green's function satisfying Neumann boundary conditions for an enclosure, which is generally constructed by slowly convergent modal expansions. The main objective of the present work is to provide an evanescent Green's function to improve the convergence of calculations. A brief description of the iPTF method and two sets of Green's functions for a rectangular cavity are presented firstly. In simulations, both the Green's functions are used to calculate the condition numbers of impedance matrices describing the relation between source and measurement patches, and the time cost of calculation based on the two sets of Green's functions at 450Hz is compared. Double pressure measurements are then employed as the input data instead of pressure and velocity measurements. The normal velocities of two baffled loudspeakers are reconstructed by the combination of a measurement method and a Green's function in the presence of a disturbing source in the frequency range of 50–1000Hz. In addition, the double pressure measurements are examined by an experiment. The precise identification of the sources indicates that the double pressure measurements are capable of localizing sources in a noisy environment. It is also found that the reconstruction with the evanescent Green's functions is slightly better than that with the modal expansions.

Sound source identification and sound radiation modeling in a moving medium using the time-domain equivalent source method.

Planar near-field acoustic holography has been successfully extended to reconstruct the sound field in a moving medium, however, the reconstructed field still contains the convection effect that might lead to the wrong identification of sound sources. In order to accurately identify sound sources in a moving medium, a time-domain equivalent source method is developed. In the method, the real source is replaced by a series of time-domain equivalent sources whose strengths are solved iteratively by utilizing the measured pressure and the known convective time-domain Green's function, and time averaging is used to reduce the instability in the iterative solving process. Since these solved equivalent source strengths are independent of the convection effect, they can be used not only to identify sound sources but also to model sound radiations in both moving and static media. Numerical simulations are performed to investigate the influence of noise on the solved equivalent source strengths and the effect of time averaging on reducing the instability, and to demonstrate the advantages of the proposed method on the source identification and sound radiation modeling.

Auditory and visual localization accuracy in young children and adults

ObjectiveThis study aimed to measure and compare sound and light source localization ability in young children and adults who have normal hearing and normal/corrected vision in order to determine the extent to which age, type of stimuli, and stimulus order affects sound localization accuracy. MethodsTwo experiments were conducted. The first involved a group of adults only. The second involved a group of 30 children aged 3 to 5 years. Testing occurred in a sound-treated booth containing a semi-circular array of 15 loudspeakers set at 10° intervals from −70° to 70° azimuth. Each loudspeaker had a tiny light bulb and a small picture fastened underneath. Seven of the loudspeakers were used to randomly test sound and light source identification. The sound stimulus was the word “baseball”. The light stimulus was a flashing of a light bulb triggered by the digital signal of the word “baseball”. Each participant was asked to face 0° azimuth, and identify the location of the test stimulus upon presentation. Adults used a computer mouse to click on an icon; children responded by verbally naming or walking toward the picture underneath the corresponding loudspeaker or light. A mixed experimental design using repeated measures was used to determine the effect of age and stimulus type on localization accuracy in children and adults. A mixed experimental design was used to compare the effect of stimulus order (light first/last) and varying or fixed intensity sound on localization accuracy in children and adults. ResultsLocalization accuracy was significantly better for light stimuli than sound stimuli for children and adults. Children, compared to adults, showed significantly greater localization errors for audition. Three-year-old children had significantly greater sound localization errors compared to 4- and 5-year olds. Adults performed better on the sound localization task when the light localization task occurred first. ConclusionsYoung children can understand and attend to localization tasks, but show poorer localization accuracy than adults in sound localization. This may be a reflection of differences in sensory modality development and/or central processes in young children, compared to adults.

Extension Method of Spherical Harmonics Beamforming for Sound Source Identification

Beamforming with a spherical array is an effective 3D sound source identification technology. The existed spherical harmonics beamforming can locate sound sources accurately, but it fails to quantify the acoustical contributions of sound sources, and that its results are affected by the focus distance seriously. Usually, pressure contribution each sound source makes to the array centre is an effective basis in ranking sound sources. In order to calculate the pressure contribution, an extension method of spherical harmonics beamforming is proposed and the influence rule of focus distance on the accuracy of the method is explored. Simulations of the given sound sources show that the method can not only locate sound sources but also calculate sound pressure contributions accurately. Concretely, the maximum error is less than 0.5 d B up to 3 000 Hz and less than 1.5 d B in 3 000-6 000 Hz. In addition, the accuracy of the method is hardly affected by the focus distance. Finally, experiment with loudspeaker sound sources validates correctness of the simulations and effectiveness of the method in practical applications.

Sound source localization identification accuracy: Bandwidth dependencies

Sound source localization accuracy using a sound source identification task was measured in the front, right quarter of the azimuth plane as rms (root-mean-square) error (degrees) for stimulus conditions in which the bandwidth (1/20 to 2 octaves wide) and center frequency (250, 2000, 4000 Hz) of 200-ms noise bursts were varied. Tones of different frequencies (250, 2000, 4000 Hz) were also used. As stimulus bandwidth increases, there is an increase in sound source localization identification accuracy (i.e., rms error decreases). Wideband stimuli (>1 octave wide) produce best sound source localization accuracy (~6°-7° rms error), and localization accuracy for these wideband noise stimuli does not depend on center frequency. For narrow bandwidths (<1 octave) and tonal stimuli, accuracy does depend on center frequency such that highest accuracy is obtained for low-frequency stimuli (centered on 250 Hz), worse accuracy for mid-frequency stimuli (centered on 2000 Hz), and intermediate accuracy for high-frequency stimuli (centered on 4000 Hz).

Identification of sound sources in soundscape using acoustic, psychoacoustic, and music parameters

This paper explores the possibility of automatic identification/classification of environmental sounds, by analyzing sound with a number of acoustic, psychoacoustic, and music parameters, including loudness, pitch, timbre, and rhythm. The sound recordings of single sound sources labeled in four categories, i.e., water, wind, birdsongs, and urban sounds including street music, mechanical sounds and traffic noise, are automatically identified with machine learning and mathematic models, including artificial neural networks and discriminant functions, based on the results of the psychoacoustic/music measures. The accuracies of the identification are above 90% for all the four categories. Moreover, based on these techniques, identification of construction noise sources from general urban background noise is explored, using the large construction project of London Bridge Station redevelopment as a case study site.

Extending interior near-field acoustic holography to visualize three-dimensional objective parameters of sound quality

It is essential to understand that the ultimate goal of interior noise control is to improve the sound quality inside the vehicle, rather than to suppress the sound pressure level. Therefore, the vehicle interior sound source localization and identification should be based on the contributions of sound sources to the subjective and/or objective parameters of sound quality at targeted points, such as driver’s ear positions. This talk introduces the visualization of three-dimensional objective parameters of sound quality based on interior near-field acoustic holography (NAH). The methodology of mapping three-dimensional sound pressure distribution, which is reconstructed based on interior NAH, to three-dimensional loudness is presented. The mathematical model of loudness developed by ANSI standard is discussed. The numerical interior sound field, which is generated by vibrating enclosure with known boundary conditions, is employed to validate the methodology. In addition, the accuracy of reconstruction of loudness distribution is examined with ANSI standard and digital head. It is shown that the results of sound source localization based on three-dimensional loudness distribution are different from the ones based on interior NAH.

Extension of planar nearfield acoustic holography for sound source identification in a noisy environment

Planar nearfield acoustic holography (NAH) is extended to identify the sound source in a noisy environment. The extended method requires the knowledge of the pressures on two closely spaced parallel hologram planes and the plane wave reflection coefficient on the target source surface. First, the incoming field coming from the back side of the microphone array and the scattered field due to the incoming wave falling on the target source are correlated through the plane wave reflection coefficient on the target source surface. Then, the mixed field on the hologram plane can be represented by the field that would be radiated by the target source into free space and the incoming field. Finally, the field that would be radiated by the target source into free space can be extracted by using the pressures measured on two hologram planes, which will be further used to accurately identify the sound source via planar NAH. The validity of the proposed method is demonstrated by simulations and experiment, and the influence of the relative strength of the disturbing source to the target source is also investigated.

Sound Source Identification and Acoustic Contribution Analysis Using Nearfield Acoustic Holography

The main goal of the present paper is to provide a method of source identification. Firstly, statistically optimal near-field acoustical holography (SONAH) techniques are applied to locate sound sources with the reflected sound field. In the presence of reflection plane parallel and perpendicular to the source plane, the incoming wave and reflected waves are separated based on the acoustic superposition principle and acoustic mirror image principle to satisfy the condition of the sound sources reconstruction using SONAH. Secondly, contribution of noise source to the special field point is analyzed and noise source ranking of interior panel groups are evaluated based the proposed three step acoustic contribution method. Finally, this method is verified experimentally.

The effects of spectral fine structure and amplitude envelope on perceived similarity and identification of approaching vehicles

The recognition and identification of important complex natural sound sources is a critical component of maintaining environmental awareness. The properties of sound sources are not static; environmental factors can cause dramatic shifts in acoustic structure. The present study investigates the effects of large imposed acoustic changes on listener perception of approaching ground and aerial vehicles. In the first experiment, listeners rated the perceived similarity of the vehicle stimuli that were either (1) unaltered, or had (2) spectral fine structure replaced with random noise, or (3) had their natural amplitude envelopes flattened. The results of a multidimensional scaling (MDS) analysis found that the ordering of perceptual space for each of the conditions was generally a function of the acoustic information available. The second experiment investigated listener identification for the same sets of sounds. Identification performance across vehicle sounds had good agreement with similarity mappings and was best when spectral fine structure was smoothed and variations in amplitude envelope were maintained, and worst when the opposite was true. These results are consistent with the notion that amplitude envelope is a critical component in the recognition and identification of complex natural sound sources.

Blind separation of sound sources from the principle of least spatial entropy

The aim of the paper is to offer a method for separating incoherent and compact sound sources which may overlap in both the space and frequency domains. This is found of interest in acoustical applications involving the identification and ranking of sound sources stemming from different physical origins. The principle proceeds in two steps, the first one being reminiscent to source reconstruction (e.g. as in near-field acoustical holography) and the second one to blind source separation. Specifically, the source mixture is first expanded into a linear combination of spatial basis functions whose coefficients are set by backpropagating the pressures measured by an array of microphones to the source domain. This leads to a formulation similar, but no identical, to blind source separation. In the second step, these coefficients are blindly separated into uncorrelated latent variables, assigned to incoherent “virtual sources”. These are shown to be defined up to an arbitrary rotation. A unique set of sound sources is finally recovered by searching for that rotation (by conjugate gradient descent in the Stiefel manifold of unitary matrices) which maximizes their spatial compactness, as measured either by their spatial variance or their spatial entropy. This results in the proposal of two separation criteria coined “least spatial variance” and “least spatial entropy”, respectively. The same concept of spatial entropy, which is central to the paper, is also exploited in defining a new criterion, the entropic L -curve, dedicated to determining the number of active sound sources. The idea consists in considering the number of sources that achieves the best compromise between a low spatial entropy (as expected from compact sources) and a low statistical entropy (as expected from a low residual error). The proposed methodology is validated on both laboratory experiments and numerical data, and illustrated on an industrial example concerned with the ranking of sound sources on a Diesel engine. At the same time, its robustness to the estimated number of active sources is demonstrated. • A method is proposed to blindly separate sound sources (i.e. without references). • Sources are assumed mutually incoherent and compact in space. • Two criteria, of least spatial entropy and of least spatial variance, are proposed. • The number of active sound sources is determined from an entropic L -curve. • The blind source separation method demonstrates good robustness.

W051002 空力音源探査のためのファン表面圧力変動計測技術の開発([W05100]流体工学部門企画,先端熱流体計測の計算との連携を含む新展開)

W051002 空力音源探査のためのファン表面圧力変動計測技術の開発([W05100]流体工学部門企画,先端熱流体計測の計算との連携を含む新展開)

Auditory informational masking and the Ear Club connection

Erv Hafter has always made a special effort to promote recognition of young researchers in our field. His Ear Club of 40-plus years has provided a conversant audience for new investigators to test new ideas and his attendant hospitality has helped forge relationships valued throughout one’s career. In this talk, I will describe how my own work benefited from visits with Erv. The focus is on informational masking (IM). I will present recent work suggesting a strong connection between two major, but seemingly unrelated, factors associated with IM: masker uncertainty and target-masker similarity. Experiments involving multitone pattern discrimination, multi-talker word recognition, sound-source identification, and sound localization are described. In each case, standard manipulations of masker uncertainty and target-masker similarity (including the covariation of target-masker frequencies) are found to have the same effect on performance provided they produce the same change in the information divergence of target and masker, a measure of statistical separation between signals from information theory. The results seem to reflect a general perceptual principle that segregates signals based on differences in their statistical structure. Future plans are to test the generality of this result in a simulated-cocktail-party environment [Hafter et al. Intl. Symp. Hearing, in press].

The information-divergence hypothesis of informational masking.

In recent years there has been growing interest in masking that cannot be attributed to interactions in the cochlea-so--called informational masking (IM). Similarity in the acoustic properties of target and masker and uncertainty regarding the masker are the two major factors identified with IM. These factors involve quite different manipulations of signals and are believed to entail fundamentally different processes resulting in IM. Here, however, evidence is presented that these factors affect IM through their mutual influence on a single factor-the information divergence of target and masker given by Simpson-Fitter's da [Lutfi et al. (2012). J. Acoust. Soc. Am. 132, EL109-113]. Four experiments are described involving multitone pattern discrimination, multi-talker word recognition, sound-source identification, and sound localization. In each case standard manipulations of masker uncertainty and target-masker similarity (including the covariation of target-masker frequencies) are found to have the same effect on performance provided they produce the same change in da. The function relating d(') performance to da, moreover, appears to be linear with constant slope across listeners. The overriding dependence of IM on da is taken to reflect a general principle of perception that exploits differences in the statistical structure of signals to separate figure from ground.

Evaluation of a separation method for source identification in small spaces

This paper investigates the efficiency of a field separation method for the identification of sound sources in small and non-anechoic spaces. When performing measurements in such environments, the acquired data contain information from the direct field radiated by the source of interest and reflections from walls. To get rid of the unwanted contributions and assess the field radiated by the source of interest, a field separation method is used. Acoustic data (pressure or velocity) are then measured on a hemispheric array whose base is laying on the surface of interest. Then, by using spherical harmonic expansions, contributions from outgoing and incoming waves can be separated if the impedance of the tested surface is high enough. Depending on the probe type, different implementations of the separation method are numerically compared. In addition, the influence of the walls' reflection coefficient is studied. Finally, measurements are performed using an array made-up of 36 p-p probes. Results obtained in a car trunk mock-up with controlled sources are first presented before reporting results measured in a real car running on a roller bench.

Half Space Acoustic Problems Analysis by Fast Multipole Boundary Face Method

Abstract: In this paper, a half space adaptive fast multipole boundary facemethod (FMBFM) is presented for solving the three-dimensional half space ex-terior acoustic problems. In the presented method, the Burton-Miller equationbased on the conventional boundary integral equation (CBIE) and its hyper-singularboundary integral equation (HBIE) is used to deal with the fictitious eigenfrequen-cies problem. The half space Green’s function is employed, thus the tree structurein the fast multipole method can be used only for the real domain. The higher orderelements and an adaptive tree structure are used to improve the efficiency of theFMBFM. This half space adaptive FMBFM for half space acoustic problems is anextension of the adaptive FMBFM for full space acoustic problems developed bythe authors. Numerical examples for half space acoustic problems in this paperdemonstrate the efficiency and validity of this method.Keywords: fast multipole boundary face method, Burton-Miller equation, acous-tic problems, half space, modified hyper-singular boundary integral equation1 IntroductionAcoustic problems in half space are a major subject in engineering. Many researchworks have been done on this aspect (Seybert et al., 1988; Secgin et al., 2010; Caoet al., 2010). The half space Green’s function (Seybert et al., 1988) can be appliedby adding the source solution at the image point to the original full-space Green’sfunction for dealing with half space problems. Sound source localization and iden-tification of a refrigerator were accomplished by using the half space boundaryelement method (BEM) (Secgin et al., 2010). Using the half space Green’s func-tion, the discretization of the rigid infinite plane is removed and only the boundaries