Map task dialogs in noise : a paradigm for examining Lombard speech

This paper presents a paradigm for comparing auditory-visual map task dialogs produced in silence and in noise, also known as Lombard speech. A previously developed temporal filtering algorithm which removes the ambient noise from recordings of Lombard speech by locating and subtracting a recording of the noise performed in the same environment was modified to accommodate longer recordings. The filtering algorithm yields overall noise attenuation between 15 and 35 dB without distorting the speech signal like spectral filtering approaches. On a small production dataset of two levels of vehicle and babble noise we examined the effect on fundamental frequency and intensity contours. We found that Lombard characteristics of speech, that is, an increase in mean F0 as well as intensity, are stronger for babble than for vehicle noise. There are indications that talkers become habituated to the noisy environment when they are exposed to it for the duration of a dialog. We did not find any consistency regarding the speed of completion of the map task, although participants appeared to solve the task more leisurely in silence than in noise. By performing eye-tracking on one of the talkers' data we found that the frequency of gaze was more than double in babble noise than in silence