The Lombard advantage (keyword score for Lombard speech minus keyword score for normal speech) in percentage points as a function of SNR and inducer noise level. The rightmost data points indicate native Lombard advantages in the −9 dB condition, taken from Lu and Cooke (2008). Absolute keyword scores for normal speech are 93.5% (quiet), 81.5% (0 dB), 62.0% (−5 dB), and 36.2% (−9 dB). Error bars here and elsewhere indicate standard errors.
Log magnitude spectra for the speech-shaped noise maskers used in experiments I and II. In each case, spectra were calculated using a 50-pole linear predictor fit to a 30 s segment of noise. Differences between the spectra SSN_normal and experiment I are due to the use of different subsets of speech material.
Mean keywords correctly identified by non-native listeners in normal and Lombard speech in speech-shaped noise maskers derived from normal and Lombard speech.
Keyword identification scores for individual talkers in quiet (upper) and combined across the two maskers (lower).
Article metrics loading...
Full text loading...