Joseph Nannariello

The prediction of reverberation time using neural network analysis

Abstract An alternative method of predicting the reverberation time for enclosures, using artificial neural networks, has been investigated. The study begins with the hypothesis that, at the conceptual design stage, reverberation time predictions are too difficult to derive using existing computer models, and too inaccurate when using other methods and that a more explicit and quicker method of predicting reverberation time is required. Assessments are made of the predictive powers of trained neural networks by comparing the predicted reverberation times obtained using neural networks with those using Sabines and Eyrings ‘classical equations’ and the ray tracing model ODEON 2.6D. The results indicate that there is a good basis for using trained neural networks to predict the reverberation time for enclosures but that 15 input variables are required to achieve accuracy of ±10%. In addition, the results show that neural network analysis can identify those variables which have the greatest effect on the predicted reverberation times.

Neural network predictions of speech levels in university classrooms

Abstract At the schematic design stage of a classroom there is a need for an expeditious and accurate method of predicting the distribution of sound levels (speech levels). The objective of this work is to investigate the possibility of developing a method of predicting the sound propagation (SP) in university classrooms, using artificial neural networks. Constructional and acoustical data for 34 randomly chosen unoccupied University of British Columbia (UBC) classrooms were used for the neural network analyses. One source position, both directional and omnidirectional sources, and a number of listener positions were chosen in each classroom making a combination of 182 cases available to train the neural networks. Assessments have been made of the method by comparing the predicted sound propagation obtained using neural networks with measured values, with predictions made using Barrons revised theory and the Hopkins–Stryker equation. The results indicate that there is a good basis for using trained neural networks to predict the distribution of sound levels in empty classrooms. The results also indicate that neural networks trained with variables which have a causal relationship to the acoustical quality of the UBC classrooms produce reliable and accurate predictions. RMS errors for Sound Propagation, in each of the frequency bands, are within the subjective difference limen for steady-state sound pressure levels, which is about 1dB (i.e. Δ E E=0.26 where E is energy density).

Introduction to neural network analysis and its application to building services engineering

The major objective of this paper is to illustrate how neural network analysis techniques might play an important role in modeling, predicting, and estimating in building services engineering. The paper outlines an understanding of how neural networks operate by way of presenting a focused discussion and illustration of appropriate problems in the different disciplines of architectural and building acoustics, civil and structural engineering and architectural science to which neural network analyses have been applied. Results of these neural network procedures, some of which are presented in this paper, are testimony to the potential of neural networks as a design tool in many areas of endeavor, including building services engineering.

A Neural-computation Method of Predicting the Early Interaural Cross-correlation Coefficient (IACCE3) for Auditoria

Abstract A method of predicting the early interaural cross-correlation coefficient (IACCE3) in unoccupied concert halls has been investigated using neural network analysis. Constructional and acoustical data for 36 unoccupied concert halls, in various countries, were utilized for the neural network analyses. A neural network for calculating IACCE3 has been embedded in a standard spreadsheet application so that designers and researchers, without access to specialized neural network software can use the results of the present work. Investigations using the neural network model have shown that IACCE3 predictions are within the subjective difference limen, which is 0.075±0.008. Five concert halls were used to assess the neural network analysis method and the errors between measured and predicted (1−IACCE3) ranged from −0.05 to 0.02. These results indicate that there is a good basis for using trained neural networks to predict IACCE3.

The use of neural network analysis to predict the acoustic performance of large rooms Part I. Predictions of the parameter G utilizing numerical simulations

Abstract A method of predicting the G values (the strength factor in dB) in large enclosures, using artificial neural networks, has been investigated. ODEON 3.1 was used to determine the acoustical attributes (including G values) for 110 unoccupied ‘shoebox’ enclosures. One source position and a combination of receiver positions were chosen, and the acoustical quantities were calculated for the 125, 250, 500, 1000 and 2000 Hz octave bands. Neural networks were then trained, verified, and tested using this data together with the size and geometrical proportions of the room. Assessments have been made of the method by comparing the predicted G obtained using neural networks with those calculated using the hybrid ray tracing computer model ODEON 3.1 for enclosures not used in the original training process. The results were also compared with Barron’s revised theory. Predictions of G in this way were in good agreement with verification and test data and Barron’s revised theory predictions. Neural networks trained with data obtained from only a small number of enclosures but with a large number of receiver positions also produced good results. The accuracy of the results for the position-dependent G, at each frequency band, is within the subjective difference limen of the acoustical parameter G, which is ±1 dB. Finally, the results show that neural network predictions based on ray-tracing programs have the potential for finding patterns in the data which could be used to establish rules of thumb to be used in the early stages of a design.

A neural network analysis of the effect of geometric variables on concert hall G values

Abstract This paper deals with the schematic design of large rooms (concert halls) using artificial neural networks. Previously published research has shown that neural networks can be used to predict the acoustical attributes ( G , C 80 , LF and RT 60 ) of concert halls. Neural network analyses have been undertaken, and the changes in neural network predictions of the parameter strength factor, G , according to the variations of concert hall design variables, has been investigated. The results of the analyses are difficult to describe simply as there appears to be a non-linear relationship between some of the many input variables. Seven concert halls have been used to show how the change in certain geometric variables will influence G values.

The Prediction of Reverberation Time Using Optimal Neural Networks

A neural network approach to predicting the reverberation time, RT60, at the conceptual design stage of auditoria, and churches is presented. The results of investigations previously carried out indicated that there was a good basis for using trained neural networks to predict the reverberation time for unoccupied enclosures but that 15 input variables were required to achieve the desired accuracy. As the number of input variables that can be readily identified and quantified at the early design stage is small, the objective of this work is to reduce network size and to obtain optimal neural networks. The results showed that the generalization performance of neural networks with simplified internal representation is efficient. Generally, the reverberation time prediction accuracy of the network models, for the six enclosures ‘tested’, is within the range of the subjective difference limen (ΔT/T ≈ 5%).

Reverberation time predictions using neural network analysis

The study aims to utilize neural network analysis to develop a method to predict the reverberation time for enclosures—in the low and mid frequencies—by using neural networks that have been trained with constructional and acoustical data. The study begins with the hypothesis that, in practice, reverberation time predictions are too difficult to undertake using existing computer models, and too inaccurate when using other methods. Specifically, the study aims at providing an expeditious and accurate method of predicting the reverberation time of enclosures at the initial design stage. To substantiate the hypothesis, and to bring into effect the aims of the study, assessments are made of the predictive powers of the trained neural networks. The results of the investigations have indicated that there is a good basis for using trained neural networks to predict the reverberation time for enclosures. The results have also shown that neural network analysis can identify those variables that have an effect on th...