11th Speech in Noise Workshop, 10-11 January 2019, Ghent, BE

Differences in pitch in native and non-native speech in noise

Katherine Marcoux(a), Mirjam Ernestus
Centre for Language Studies, Radboud University, Nijmegen, The Netherlands

(a) Presenting

When speaking in noise, the acoustics of our speech change and we produce Lombard speech. Extensive research on native speakers has shown that, compared to speech produced in quiet, Lombard speech has, among other properties, a higher Fundamental Frequency (F0, pitch), higher amplitude (loudness), and a decrease in spectral tilt (shift in energy) (e.g. Summers et al. 1988). Lombard speech is typically viewed as a reflex, shown by speakers across different languages. Non-natives, however, are confronted with several added difficulties when producing Lombard speech, including experiencing a higher cognitive load and potential differences in pitch ranges between the two languages (Jenner 1976; Wen, Mota, and McNeill 2015). Considering these added challenges, we wanted to investigate whether native and non-native Lombard speech differs in terms of F0.

We recorded 30 Dutch natives reading stimuli in Dutch and English and 9 American-English natives in English, in quiet and noise (hearing 83 dB SPL Speech-Shaped Noise through headphones). We additionally manipulated the focus in the sentences by means of contrastive question answer pairs, creating early and late focus answer sentences. This resulted in 36 question-answer pairs in four conditions (quiet early-focus, quiet late-focus, noise early-focus, and noise late-focus), for instance, “No, the tall (focus) woman drove to the pub in town” (early-focus) and “No, she drove to the pub (focus) in town.” (late-focus).

The recordings were segmented at the sentence level and Praat returned the median F0 value. Using the lme4 package in R, we conducted linear mixed effects models with language or nativeness, noise, and focus as fixed effects and random slopes for participant and stimuli as random effects. Our preliminary analyses illustrated that there was an effect of noise for the Dutch speakers, leading to an increase in F0 in their production of English and Dutch Lombard speech as compared to their speech produced in quiet. This effect of noise was only present for the American-English speakers in the late-focus condition.

The data on native and non-native Lombard speech suggest that non-native speakers also produce Lombard speech, supporting the hypothesis that Lombard speech is a reflex. Against the reflex hypothesis, however, are our findings that, rather than being a universal phenomenon, Lombard speech interacts with the phonetic properties of the language. This indicates that it is more language specific than originally expected and that it has to be acquired by second language learners.

Last modified 2019-01-08 16:51:41