Using SIL/PSIL to estimate Speech Intelligibility in Vehicles

The paper looks for the degree of linear correlation of some acoustic parameters concerning Speech Intelligibility in vehicles. The Articulation Index (AI) was originally a criterion to characterize the influence of parasite noise on the Intelligibility of a conversation in the design of speech communication systems. Introduced in the automobile acoustics, it is more and more commonly used by vehicles manufacturers to estimate the middle and high frequency content of spectra noise inside various types of vehicles driven under several running conditions. The correlation with Speech Intelligibility, measured through subjective measurements is well known. Presently we have more sophisticated parameters like STI, RASTI, SII… directed mainly to architectural acoustics and sometimes used in vehicle acoustics. They require specific hardwares and softwares and are more complicated to deal with and often are called machine measures of Speech Intelligibility. On the other side, a simplified way to estimate Speech Intelligibility was proposed by Beranek in the fifties, introducing Speech Interference Level (SIL). Later a new version named Preferred Speech Interference Level (PSIL) was developed. They are calculated using measurements of sound pressure levels in octave bands above 500 Hz. This work, as a pioneer application, uses a set of passenger cars and a set of commercial vehicles to study the correlation of these parameters regarding internal noise. Besides the usual definitions, we propose new approaches for these alternative parameters, SIL and PSIL, introducing some frequency weightings similar to those existing in the calculation of AI in third octave bands. We apply to the experimental data a least-squares curve fitting by straight lines. The results observed through graphics show the degree of linear correlation for the separated sets and also for all the data jointed together. For some parameters we obtain a high degree of linear correlation in such way we expect simplifications and facilities in the estimation of Speech Intelligibility inside vehicles. We use propagation of errors in the mathematical expressions regarding the acoustic parameters to discuss the establishment of a ranking using a specific category of vehicles. We can see that using sound pressure levels in octave bands from 1kHz to 4kHz we are able to have SIL as an alternative picture of Speech Intelligibility, and due to this simplicity we also recommmend them for quality control.