A multichannel learning-based approach for sound source separation in reverberant environments

In this paper, a multichannel learning-based network is proposed for sound source separation in reverberant field. The network can be divided into two parts according to the training strategies. In the first stage, time-dilated convolutional blocks are trained to estimate the array weights for beamforming the multichannel microphone signals. Next, the output of the network is processed by a weight-and-sum operation that is reformulated to handle real-valued data in the frequency domain. In the second stage, a U-net model is concatenated to the beamforming network to serve as a non-linear mapping filter for joint separation and dereverberation. The scale invariant mean square error (SI-MSE) that is a frequency-domain modification from the scale invariant signal-to-noise ratio (SI-SNR) is used as the objective function for training. Furthermore, the combined network is also trained with the speech segments filtered by a great variety of room impulse responses. Simulations are conducted for comprehensive multisource scenarios of various subtending angles of sources and reverberation times. The proposed network is compared with several baseline approaches in terms of objective evaluation matrices. The results have demonstrated the excellent performance of the proposed network in dereverberation and separation, as compared to baseline methods.

Efficient prediction of construction equipment exterior and cabin interior noise over broad frequency range using novel SEA method

Statistical Energy Analysis (SEA) is commonly used for the prediction of interior cabin noise from construction equipment such as excavators, dump trucks, or graders. While traditional SEA method is computationally efficient and effective for the prediction of total radiated noise, it isn't suitable for prediction of sound diffraction around machinery and evaluation of spatial variations in sound field. As a result, prediction of cabin airborne interior noise transmission using SEA method typically requires experimental measurements in order to estimate incident sound field over the exterior boundary of the cab which makes it unsuitable for use in early stage design where test data isn't available. A novel SEA method that accounts for spatial gradients in the reverberant field has been developed and is introduced in this paper. It's usage for prediction of both exterior and cab interior noise over broad frequency range is demonstrated along with experimental validation for construction equipment under operating conditions.

Near-Direct and Far-Reverberant Field Response of Direct Radiator Loudspeakers

Typical audience seating arrangements in rooms and auditoria warrant reinvestigation of the direct radiator speaker response in the near-direct and far-reverberant fields, as the response data provided by the manufacturer is always ideal and does not account for the effect of those fields. The speaker response characteristics of a variety of direct radiator loudspeakers ranging from the conventional squawker to the full range radiator have been investigated in these fields. The speaker response is investigated in the 50 -10 kHz frequency range, by measuring the A-weighted SPL (sound pressure level) in the near-direct and far-reverberant fields, using an acoustic analyzer. The field-specific characteristic for each of the radiators is determined by fitting the SPL data obtained to an appropriate polynomial. The coefficients obtained thereby, allow an objective field-specific study amongst radiators. When a set of direct radiator loudspeakers is available, it is necessary to configure their application, depending upon the optimum sound quality required for a given enclosure, in near-direct field and far-reverberant field. The outcome of this work assists one to configure the best radiator ensemble for a given enclosure, despite placement constraints.

An Investigation Into Acoustic Dipole Source Calibration Methods for Turbomachinery Sound Radiation in Reverberant Fields

In-situ calibration methods using a single spherical-shaped transmitting hydrophone (idealized as a monopole acoustic source) have traditionally been used for radiated sound measurements of turbomachinery performed in the Garfield Thomas 1.22-m diameter water tunnel located at The Pennsylvania State University’s Applied Research Laboratory (ARL Penn State). In this reverberant field, the monopole source containing known transmitting characteristics was used to calibrate acoustic sensors that were either near or far from the source. This method typically works well when the type of source is monopole in nature; however, many acoustics sources can be dipole or quadrupole in nature. In this study we investigated the applicability of using dipole sources in a space such as a well-characterized reverberant tank, and we found through a virtual dipole method that the radiation still appears monopole in the reverberant field. The method was extended for the vibration of a panel (a known dipole source) and once again the monopole assumption for the in-situ calibration for a near-field hydrophone and conventional reverberant hydrophones remained consistent.

Commissioning of an Atypical Acoustic Facility for Experimental Testing

Computational modeling (BEM, FEM, and SEA) is often implemented at different stages of the design process to optimize manufacturing and performance parameters. Computational results are typically verified experimentally. Experimental testing standards, particularly those related to vibro-acoustic testing, are defined by various agencies such as ASTM, ANSI, and ISO. An investigation proposing a new computational methodology of analyzing the vibro-acoustic behavior of an aircraft fuselage due to the turbulent boundary layer required verification of the predictions experimentally. In the face of certain limitations, an atypical acoustic facility was constructed challenging conventional standards while complying with the defined criteria of international testing standards. Principal deviations relate to the geometric requirements that recommend large volumes of certain construct, and microphone and acoustic source positioning. The calculated 95% confidence intervals compared exceptionally well against defined criteria (strictest measure is 1 for frequencies greater than 315 Hz) by averaging less than 0.4 for each test product across a frequency range that exceeded is the range specified by ASTM E90. The requirements for qualification of the reverberation chamber according to ANSI S12.51 were also satisfied, with the exception of measurements at 125 Hz and 160 Hz that observed heightened sensitivity due to near field effects and room modes. The calculated permissible ratio of decay variation showed good agreement against ASTM C423 criteria despite the intrinsic challenge of creating a diffuse and reverberant field in a confined, or constricting, volume. The last compliance measure reviewed flanking to ensure acceptable signal-to-noise ratio. It was clearly demonstrated that the silenced sound pressure levels (with the presence of the specimen) were greater than 10 dB above the background sound pressure levels (with the consequences of flanking considered). The investigation confirmed the feasibility of using an atypical acoustic facility to comply with various international testing standards. The noted deviations and shortcomings are not specific to the presented work, but are common challenges that all facilities observe.

Sensitivity Analysis of Acoustic Field Parameters on a Change of Boundary Conditions in a Room

The area of environmental protection concern minimises the impact that technical objects have on the environment. Usually the most effective way of protecting the environment is to influence the source of the problem. For this reason studies are conducted to modify the construction of machines, power machines in particular, so as to minimise their impact on the environment. In the case of environmental protection from noise it is most convenient to carry out measurements in an anechoic chamber. Unfortunately, this is possible only in very limited circumstances. In all other cases measurements are performed using an engineering method or the survey method, both of which are described in the standards and by taking into account the so-called environmental corrections. The obtained results are burdened with greater error than those of measurements in an anechoic chamber. Therefore, it would seem advantageous to develop a method of obtaining similar and reliable results as those in an anechoic chamber, but in a reverberant field. The authors decided to use numerical modelling for this purpose. The main objective of this work is a comprehensive analysis of the numerical model of a laboratory designed for acoustic tests of selected power machines. The geometry of a room comprising an area of analysis is easy to design. The main difficulty in modelling the phenomena occurring in the analysed area can be the lack of knowing the boundary conditions. Therefore, the authors made an attempt to analyse the sensitivity of various acoustic parameters in a room in order to change these boundary conditions depending on the sound absorption coefficient