Reverberation Predictions Research Articles

Speech reverberation suppression for time-varying environments using weighted prediction error method with time-varying autoregressive model

In this paper, a novel approach for the task of speech reverberation suppression in non-stationary (changing) acoustic environments is proposed. The suggested approach is based on the popular weighted prediction error (WPE) method, yet, instead of considering fixed reverberation prediction weights, our method takes into account the more generic time-varying autoregressive (TV-AR) model which allows dynamic estimation and updating for the prediction weights over time. We use an initial estimate of the prediction weights in order to optimally select the TV-AR model order and also to calculate the TV-AR coefficients. Next, by properly interpolating the calculated coefficients, we obtain the ultimate estimate of reverberation prediction weights. Performance evaluation of the proposed approach is shown not only for fixed acoustic rooms but also for environments where the source and/or sensors are moving. Our experiments reveal further reverberation suppression as well as higher quality in the enhanced speech samples in comparison with recent literature within the context of speech dereverberation.

Effects of time dispersion on echo, reverberation, and echo to reverberation ratio in a range-dependent Pekeris waveguide

In shallow water, active sonar performance is typically limited by reverberation, making the prediction of target echo and reverberation, and their ratio, an important part of sonar performance prediction. In range-dependent shallow water environments, these quantities are often calculated without considering the effect of dispersion. The effect of time dispersion is considered taking examples for a range-dependent bathymetry from the 2010 Weston Memorial Workshop. Using the analytical method of [M. A. Ainslie and D. D. Ellis (in press), IEEE Journal of Oceanic Engineering], combined with normal mode predictions [D. D. Ellis (1995). The Journal of the Acoustical Society of America, 97(5), 2804-2814], neglect of time dispersion is found to result in an error of up to 16 dB in the echo level for a short CW pulse (duration 3 ms). The effect of 3D geometry is considered, and results for a cylindrically symmetric bathymetry are shown to differ by up to 14 dB from the corresponding results with a Cartesian symmetry. The difficulties associated with modeling an LFM pulse are discussed.

Echo, Reverberation, and Echo-to-Reverberation Ratio for a Short Pulse in a Range-Dependent Pekeris Waveguide

In shallow water, active sonar performance is typically limited by reverberation, making the prediction of target echo, reverberation, and echo-to-reverberation ratio an important part of sonar performance prediction. In range-dependent shallow-water environments, the echo is often calculated without considering the effect of dispersion, and reverberation predictions are often limited to Lambert's rule for seabed scattering. In this paper, analytical formulae are derived for the echo, including the effect of time dispersion, and for reverberation for non-Lambert scattering laws. More specifically, the method for calculating the short-pulse echo intensity developed by Harrison and Nielsen is extended to a range-dependent bathymetry. For reverberation, the Zhou–Harrison method is also generalized to cover an arbitrary power law for the seabed scattering coefficient, combined with a range-dependent bathymetry. These formulae are applied to range-dependent test cases from an international workshop, and compared with predictions using a normal mode sum. Neglect of time dispersion is found to result in an error of up to 16 dB. Results for a cylindrically symmetric bathymetry differ by up to 14 dB from the corresponding results with a Cartesian symmetry.

Observations of Sea-Surface Waves During the 2013 Target and Reverberation Experiment (TREX13) and Relation to Midfrequency Sonar

As part of the 2013 Target and REverberation eXperiment (TREX13), which took place off the coast of Panama City, FL, USA, directional wave measurements were made using two directional wave buoys separated in range by 5 km. The purpose of these measurements was to provide environmental support for the interpretation of reverberation and other active sonar experiments that were part of TREX13. During the measurement period between April 22 and May 17, 2013 exclusive of a period of nondeployment May 2–6, 2013, the root-mean-square (rms) wave height $H$ varied over the range 0.03–0.33 m, holding a median value of 0.11 m; the wind speed varied from ${\sim} $ 1 to 10 m/s with a median value of 4.7 m/s, and the rms wave slope averaged over all directions varied from 0.01 to 0.10 with median value of 0.05. These parameters are placed in the context of midfrequency sonar propagation and reverberation prediction. One buoy operated the entire period, with the second buoy operating simultaneously over a four-day overlap period, during which there was excellent agreement between $H$ and wave slope in two orthogonal directions, a finding relevant to describing the sea surface as spatially invariant, or homogeneous, for purposes of sonar modeling. The analysis of energy-weighted mean direction illustrates how the wave field was generally composed of a mixture of swell and wind-generated waves; in cases of purely wind-generated waves the effect of a limited fetch was also shown.

Speech dereverberation using weighted prediction error with correlated inter-frame speech components

In this paper, we propose a new dereverberation approach based on the weighted prediction error (WPE) method implemented in the short-time Fourier transform (STFT) domain. Our main contribution is to model the temporal correlation of the STFT coefficients across analysis frames, referred to as inter-frame correlation (IFC), and exploit it in the dereverberation process. Since accurate modeling of the IFC is not tractable, we consider an approximate model wherein only a finite number of consecutive speech frames are considered correlated. It is shown that, given an estimate of the IFC matrix, the proposed approach results in a convex quadratic optimization problem with respect to the reverberation prediction weights, and a closed-form solution can be accordingly derived. Furthermore, an efficient method for the estimation of the underlying IFC matrix is developed based on the extension of a recently proposed speech variance estimator. We evaluate the performance of our approach incorporating the estimated IFC matrix and compare it to the original and several variants of the WPE method. The results reveal lower residual reverberation and higher overall quality of the enhanced speech when the proposed method is employed.

A modeling approach to scattering and reverberation in shallow water

A physics-based modeling approach is described that allows prediction of reverberation in complex shallow water environments. An integral expression is presented for the backscatter intensity with a factorized integrand comprised of two kernels, the two-way propagator and the scattering kernel. The propagator is defined by Green’s function, describes the local intensity, and can be calculated using available models, such as PE, normal modes, or ray approximations. An expression for the scattering kernel is obtained using a unified approach to volume and roughness scattering [Ivakin, JASA Feb. 1998], and represents a sum of volume and roughness scattering coefficients. They are specified for sea-water column and rough heterogeneous seabed with continuous spatial fluctuations of compressibility and density, and/or discrete randomly distributed targets, such as bubbles, fish, shells, and others. Results of numerical simulations for shallow water reverberation time/range series, based on a PE propagation model, are presented, and potential contributions of different mechanisms of scattering are compared and discussed. The approach is applied to consider reverberation in a complex shelly sand/mud environment, such as one at the Target and Reverberation Experiment 2013 (TREX), and results in model/data comparisons based on analysis of TREX acoustic scattering data and environmental ground truth measurements. [Work supported by ONR-OA.]

Improved Active Sonar Tactical Support by Through-the-Sensor Estimation of Acoustic Seabed Properties

Accurate environmental information is required for obtaining confident sonar performance predictions. This environmental information is, however, often unreliable or unavailable. To support antisubmarine warfare (ASW) operations, a through-the-sensor approach has been developed in which relevant acoustic seabed properties are derived from reverberation data, and a demonstrator system has been installed on a Royal Norwegian Navy frigate. It determines relevant acoustic seabed parameters from the reverberation data near real time. This demonstrator system has been validated in several sea trials conducted off the coast of Bergen, Norway. The acoustic seabed parameters derived in these trials have a good correspondence with the available prior information. Furthermore, the results show that acoustic seabed parameters derived from reverberation data in previous trials can be used to improve reverberation prediction for subsequent trials, even when environmental conditions, i.e., sound-speed profiles, are different. Because the demonstrator makes information on acoustic seabed properties directly available for in situ sonar performance prediction, it can be used as a tactical decision aid.

An initial model-data comparison of reverberation and clutter from a near-shore site in the Gulf of Mexico

Reverberation measurements were made in the Gulf of Mexico off Panama City, Florida, USA, in April 2012 in preparation for the main Transmission Reverberation Experiment (TREX) in May 2013. The data were gathered using the triplet section of the ONR Five Octave Research Array (FORA), deployed as a fixed receiver. By steering cardioid beams to the right or left, the array can reduce ambiguity. Beamformed data from the 2012 trial show background noise with high directionality and variability due to nearby shipping. Model predictions of reverberation and target are compared with data using a range-dependent Clutter Model, which uses adiabatic normal modes as the computational engine. The initial predictions use isovelocity water, over a sandy bottom halfspace with Lambert scattering, and bathymetry from the GEBCO08 database. These initial results will be presented, hopefully supplemented by improved predictions with better environmental inputs and additional clutter data obtained during the May 2013 experiment. [Work supported by ONR Code 322 OA.]

Comments on “Empirical Prediction of Speech Levels and Reverberation in Classrooms” [Build Acoust, 8(1), 1–14, 2001

The goal of this letter is to comments on the article “Empirical Prediction of Speech Levels and Reverberation in Classrooms” by Hodgson (2001). Hodgson proposed an empirical equation for prediction of classroom early decay time. Hodgson EDT equation with regard to the classroom ceiling height was theoretically discussed and it was concluded that when a classroom ceiling height is decreased or increased, it is not reasonable to judge about the acoustical condition of the new space based on the Hodgson EDT equation if the average absorption coefficient of the additional walls is equal to the average absorption coefficient of the original space.

High-frequency broadband coherent reverberation predictions for the reverberation modeling workshop.

The reverberation modeling workshop included cases with strong attenuation to the forward propagation caused by multiple forward scattering. The effects of this attenuation were consolidated into effective reflection coefficients using the small slope approximation and provided to workshop participants. In this talk the coherent modeling of reverberation for these strongly attenuated cases using complex modes as computed by KRAKENC is described, and the results of these coherent modal predictions are compared to results obtained by other participants using ray theory. The importance of including both the complex mode shapes and eigenvalues, as well as the coherent interactions between modes, to obtain accurate predictions is discussed.

TWO MODELING APPROACHES FOR REVERBERATION IN A SHALLOW WATER WAVEGUIDE WHERE THE SCATTERING ARISES FROM A SUB-BOTTOM INTERFACE

In shallow water environments where the uppermost sediment layer is a fine-grained fabric (e.g. clay or silty-clay), the observed reverberation may be dominated by scattering from the sub-bottom. Here, reverberation predictions from normal mode and energy flux models are compared for the case where the scattering arises from a sub-bottom half-space under a fine-grained sediment layer. It is shown that in such an environment, the position of the angle of intromission, in addition to the angular dependence of the scattering kernel, is a factor controlling the reverberation and its vertical angle distribution. It is also shown that the reverberation from a sub-bottom horizon is typically governed by higher grazing angles than the case where the scattering occurs at the water–sediment interface. There was generally very close agreement between the models as a function of frequency (200–1600 Hz), layer thickness (0–8 m), and range (1–15 km). The model comparisons, showing some differences, illuminate the result of different approximations in the two approaches.

Rapid simulation of sonar time series.

Rapid simulation of element level time series for bistatic sonars requires quality approximations in order to yield accurate results in real time. Here rapid physics‐based time series approximations based on closed form expressions for reverberation intensity developed by Harrison [J. Acoust. Soc. Am. 114, 2744 (2003)] are developed for boundary reverberation and target scattering in shallow water waveguides. The simulations include Doppler and waveguide angle and time dispersion effects, and since the simulations are at the element level, they can be processed like real data. Simulations are compared to reverberation predictions obtained with more accurate but slower codes showing good agreement.

Fitting data, but poor predictions: Reverberation prediction uncertainty when seabed parameters are derived from reverberation measurements

For many decades, researchers have been developing inverse techniques for estimating seabed parameters from reverberation data, notably scattering strength. Generally, the angular dependence of the scattering kernel is unknown and is either solved for or assumed fixed. In either case, agreement is typically quite good between the measured reverberation and that modeled (by fitting scattering parameters). However, what are the resulting uncertainties in a reverberation prediction if the ocean or geometry changes? The main results of the paper are that (1) these prediction uncertainties are surprisingly large, of order 10 dB at 10 km and thus (2) traditional/current methods for reverberation inversion should be augmented, mitigating the large prediction uncertainties by an additional measurement. Several options for additional measurements are discussed.

The importance of retaining inter-path coherence in reverberation prediction

The R-SNAP and BiStaR normal mode-based reverberation models are used to solve a selection of problems from the 1st Reverberation Modeling Workshop held in Austin, TX in November of 2006. R-SNAP and BiStaR are codes that include the effects of inter-mode decorrelation to predict the coherent aspects of reverberation, as well as allowing for the option to compute the ‘‘incoherent’’ reverberation. Comparisons between the incoherent R-SNAP and BiStaR predictions and model predictions obtained by other methods show good agreement. However, it is argued here that as defined, the problems require a coherent solution, and differences between the incoherent and coherent solutions can be significant. Comparisons with coherent 2-way PE results for problem I computed by Joe Lingevitch (NRL) show that coherent structure in reverberation exists, is predictable, and should therefore be an integral part of any reverberation calculation. [Work supported by ONR.]

A comparison of ray, normal-mode, and energy flux results for reverberation in a Pekeris waveguide

A number of problems were developed for, and presented at, a 2006 Reverberation Modeling Workshop sponsored by the US Office of Naval Research. The simplest of these known to the participants as Problem 11, was a Pekeris waveguide (isospeed water over a flat bottom half-space) with Lambert bottom scattering. The water depth was 100 m and frequencies of 250, 1,000, and 3,500 Hz were specified. A number of source-receiver combinations were specified, but the reverberation predictions are quite insensitive to sensor depth except at 250 Hz. With some benefit from hindsight and the results from other models, we compare our results from ray, normal-mode, and energy-flux approaches. All three approaches agree at intermediate times, say 3 to 50 s. At short times, the steep-angle paths and fathometer returns cause the mode and energy-flux models to underpredict the reverberation. At longer times, the ray models run out of steam: i.e., there are too many contributing ray paths for them to handle so they underpredict the reverberation. By combining the model predictions together with analytical results from an energy-flux model, we propose a composite benchmark solution. [Work supported in part by ONR.]

Reverberation modeling with BiRASP—The bistatic range-dependent active system prediction model

The calculation of bistatic reverberation is a computationally intensive exercise. Efficient computation of timely results typically involves numerous physical and algorithmic assumptions or approximations. Consequently, it is not unusual for predicted time series from different models to differ in level or structure. At a recent Reverberation Modeling Workshop [9–13 November 2006, Austin, TX], a set of test cases were proposed with the intent of establishing benchmark reverberation problems and solutions. This talk will focus on the test case performance of BiRASP, the Naval Research Laboratory’s bistatic reverberation model. BiRASP is a ray-based code, originally developed for deep-water, long-range, low-frequency reverberation predictions and data analysis. Results for the various test case problems will be shown. The consequences of some of the typical algorithm and physical assumptions/approximations will be demonstrated and discussed. [Work supported by the Office of Naval Research.]

A comparison of two modeling approaches for reverberation in a shallow-water waveguide where the scattering arises from a sub-bottom interface

In shallow water environments with a low-speed sediment layer overlying a higher-speed sub-bottom, the observed reverberation may be dominated by scattering from the sub-bottom. Here, reverberation predictions from normal mode and energy flux models are compared for the case where the scattering arises from a sub-bottom half-space under the low-speed sediment layer. It is shown that in such an environment, the position of the angle of intromission, in addition to the angular dependence of the scattering kernel, is a factor controlling the vertical angle distribution. It is also shown that the reverberation from a sub-bottom horizon is typically governed by higher grazing angles than the case where the scattering occurs at the water-sediment interface. There was generally very close agreement between the models as a function of frequency (200–1,600 Hz), layer thickness (0–8 m), and range (1–15 km). The model comparisons, showing some differences, illuminate the effect of different approximations in the two approaches. [Work supported in part by the Office of Naval Research.]

Constrained comparison of ocean acoustic reverberation: Theory and observations

Measurements of long-range (order 104 m) shallow-water reverberation in the Straits of Sicily at 900 and 1800 Hz are compared with theoretical predictions. All of the required environmental inputs for the theory are obtained independently, that is to say there are no free parameters. The reflection coefficient and the scattering strength are measured by direct path methods; both quantities show strong frequency dependence. The theoretical reverberation predictions using these measurements are in good agreement with directional reverberation data, i.e., within the expected uncertainty bounds. The good agreement suggests that the supporting environmental measurement techniques are robust and that the physics associated with reverberation in a waveguide is reasonably well understood, at least in simple environments. The ability to independently measure the seabed scattering strength and reflection coefficient is a crucial step for the advancement of inverse methods using reverberation (e.g., rapid environmental assessment) inasmuch as it provides the means for quantitatively measuring the robustness of those methods. [Research supported by the Office of Naval Research OA321 and the NATO Undersea Research Centre.]

A Matlab and normal mode based reverberation model and some data/model comparisons

A Matlab and normal mode based reverberation model has been developed that uses Ellis’ algorithm [J. Acoust. Soc. Am. 97, 2804–2814] for reverberation predictions and the ORCA normal mode model [Westwood et al., J. Acoust. Soc. Am. 100, 3631–3645] to compute the eigenvalues and mode functions. The model is currently range independent, but handles bistatic geometries, and towed array beam patterns. Group velocity corrections similar to LePage’s [J. Acoust. Soc. Am. 106, 3240–3254] have recently been added to this model. The matrix formulation in Matlab allows the bistatic calculations to be performed more efficiently. An overview of the model is presented that includes sample model-model comparisons for some problems from the 2006 ONR Reverberation Workshop. The model predictions are also compared with data obtained with the NURC (formerly the NATO Undersea Research Centre, SACLANTCEN) towed arrays during several reverberation experiments on the Malta Plateau. Select data from these experiments are used to compare with model predictions using best estimates of bottom properties in the area. [Work supported by US Office of Naval Research, Code 321OA, Grant No. N00014-05-1-0156.]

Prediction of reverberation by stochastic processes in a forest: Creation of reverberation in an outdoor performance space

A forest is an outdoor space. For that reason, reverberation formulas of indoor spaces cannot be used when predicting forest reverberation. As a matter of fact, because the level decay curve of the reverberant sound produced in a forest is nonlinear, the concept of reverberation time is inapplicable. We clarified that reverberation of a forest is expressible as a Wiener process (Brownian motion), which is a stochastic process, and we derived the formula of reverberation of a forest based on that fact. In this study, Monte Carlo simulation was also carried out. Results clarified the following. (1) The energy decay curve of the reverberation sound of a forest becomes a straight line on the logarithmic scale of time and energy; its line gradient is −2. (2) At a point far from a sound source, the reverberant sound energy increases first and begins to decrease after some period. (3) When the mean‐free path is short, the sound level of reverberation becomes high. Furthermore, we devised some methods of generating reverberation in outdoor space using the forest reverberation mechanism. For example, a proposed reverberation wall can produce reverberations using only one wall incorporating sound scatterers.