Summary of potential masking effects for native and non-native listeners.
Native (N) and non-native (NN) keyword identification scores in rationalized arcsine units (RAUs) for quiet and in three levels of speech-shaped noise for color, letter, and number keywords. The native advantage (N-NN in RAUs) is also shown. Error bars here and elsewhere denote standard errors.
Effect of speaker gender (top) and speech rate (bottom) on intelligibility for native and non-native listeners in the four conditions of experiment 1. Scores are averaged over color, letter, and number keywords and transformed to RAUs. For speech rate, the ordinate represents the mean duration of the fastest and slowest thirds of the utterance set.
Intelligibility of individual talkers for native and non-native listeners in the four conditions of experiment 1. Solid lines indicate mean intelligibilities across all talkers for the two listener groups, while dotted lines locate s.d. from the mean (these are plotted where the value is within the range of talker scores). Male and female talkers are identified by m and f, respectively, and the numbers distinguish different talkers in the Grid corpus. Scores computed as for Fig. 3.
Native and non-native keyword identification scores in the two-talker conditions.
Upper panel: Effect of fundamental frequency differences in the two-talker conditions. Keyword identification scores for natives and non-natives in the three subconditions (sg=same gender, st=same talker, dg=different gender) are presented for subsets of utterance pairs in the lower and upper tercile of F0 differences. Lower panel: Effect of absolute duration on keyword identification in the two-talker conditions.
Proportions of keywords from the target utterance (black) and from the masker (mid gray). The residual (light gray) shows the proportion of responses which were not part of the target or masker.
Upper panel: Automatic speech recognition scores based on glimpse recognition for the stimuli of experiments 1 and 2. Rather than being expressed in terms of SNR (experiment 1) or TMR (experiment 2), recognition scores are plotted as a function of the mean glimpse percentage on which recognition was based. Lower panel: Solid lines (after pairwise linear interpolation between the four SNRs indicated) depict glimpse proportion vs intelligibility (scored as the mean identification rate of the letter and digit keywords) derived from experiment 1, for native and non-native listeners. Vertical lines indicate the measured glimpse percentages for the six TMR conditions of experiment 2.
Article metrics loading...
Full text loading...