19th INTERNATIONAL CONGRESS ON ACOUSTICS MADRID, 2-7 SEPTEMBER 2007 PREDICTION OF SPEECH INTELLIGIBILITY ALONG UNDERGROUND PLATFORMS USING A WEB BASED MODEL

Over the last 5 years a computer prediction model based on a web browser has been developed, www.whyverne.co.uk; first to predict sound levels in rooms, followed by early decay time and reverberation time. Now Speech Transmission Index (STI) can be calculated for many identical sound sources in any type of room with or without noise control. Comparisons of sound propagation and early decay time in a hypothetical and a scale model of an underground station were undertaken for one and many sound souces. Next, the web model predicted actual STI measurements taken on a London Underground platofrm. Finally CATT-Acoustic was used to predict the speech intelligibility on the station platform. The predictions by both models were within 0.05 STI, or 1 limen of the measured STI values along the length of the platform, although the CATT-Acoustic model was more accurate, the web model was significantly quicker at predicting the results. INTRODUCTION The aim of this research was to develop the web model, CISM, so that it was capable of predicting speech intelligibility on Underground platforms from multiple identical sound sources. CISM is freely available on-line and capable of predicting 9 receiver positions, in a room with fittings or various noise control methods [1] from up to 18 sound sources. It should be noted that source directivity of the speaker was not considered, although this is to be simulated in the next version of the model, based on the Windows model [2]. To give a comparison the commercial software CATT-Acoustic v8.0 was used to predict SPL, RT, STI and RASTI in the rooms [3]. The benefit of the CISM model is that it predicts each receiver in a few seconds, rather than a few minutes as in the case of CATT [1]. The accuracy of the models was assessed against predictions from CATT-Acoustics, and measurement taken using the MLSSA system in real spaces. CISM MODEL CISM is an image source based model, capable of modeling multiple directional sound sources in rooms with fittings near the floor and/or ceiling, with rectangular absorptive patches on any of the six walls and total absorptive rectangular barriers in the room. CISM can predict sound pressure level and reverberation time in each octave band, each band being processed separately. This model can be learnt in 5 minutes without any prompting, assuming an undergraduate level of understanding of acoustics, as established by the MSc Architectural Acoustics students. The noise control aspects of the web model take approximately 10 minutes to master, based on a sample of six students [4], see Figure 1. To predict STI it was necessary to make various amendments. To wit, simultaneous processing of all 6 octave bands, associated changes in the reflection order, introduction of background noise levels, the STI/RASTI calculation itself and the necessary room description [5]. The model uses EDT to calculate reverberation in the room. The reflection orders, one per octave band, were based on a 99% energy discontinuity [6]. The largest of these was used as the reflection order, contributions beyond each octave’s reflection order were ignored. This aspect of the web model had to be reduced to a fixed 14 reflections, as Javascript code is limited compared to native executable code.