Binaural speech-to-noise loudness ratio at the speech reception threshold in vehicles

Abstract

Traditional methods for predicting the intelligibility of speech in the presence of noise inside a vehicle, such as the Articulation Index (AI), the Speech Intelligibility Index (SII), and the Speech Transmission Index (STI), are not accurate, probably because they do not take binaural listening into account; the signals reaching the two ears can differ markedly depending on the positions of the talker and listener. We propose a new method for predicting the intelligibility of speech in a vehicle, based on the ratio of the binaural loudness of the speech to the binaural loudness of the noise, each calculated using the method specified in ISO 532-2 (2017). The method was found to give accurate predictions of the speech reception threshold (SRT) measured under a variety of conditions and for different positions of the talker and listener in a car. The typical error in the predicted SRT was 1.3 dB, which is markedly smaller than estimated using the SII and STI (2.0 dB and 2.1 dB, respectively).

Effect of Acoustic Scene Complexity and Visual Scene Representation on Auditory Perception in Virtual Audio-Visual Environments

“…Because of the very similar patterns of results for speech intelligibility and loudness, one could expect that target speaker loudness is a good predictor for speech intelligibility in multiple masker conditions. Recently, [35] successfully predicted speech intelligibility of speech in the steady background noise of a car based on the binaural speech-to-noise loudness ratio. Such metric is assumed to be strongly related to the here measured target speaker loudness in the presence of fluctuating ISTS maskers.…”

“…4 and 5) indicates that target speaker loudness seems not sensitive enough as reliable estimator for speech intelligibility in reverberation. As stated in [35] effects of reverberation are negligible for listening in vehicles, and accordingly its influence on the prediction performance of their model was not assessed so far. An increase in the number of maskers also reduced the accuracy for direction and distance estimations.…”