Recognition of accented and unaccented speech in different maskers by younger and older listeners

Abstract

This investigation examined the effect of accent of target talkers and background speech maskers on listeners' ability to use cues to separate speech from noise. Differences in accent may create a disparity in the relative timing between signal and background, and such timing cues may be used to separate the target talker from the background speech masker. However, the use of this cue could be reduced for older listeners with temporal processing deficits, especially those with hearing loss. Participants were younger and older listeners with normal hearing and older listeners with hearing loss. Stimuli were IEEE sentences recorded in English by male native speakers of English and Spanish. These sentences were presented in different maskers that included speech-modulated noise and background babbles varying in talker gender and accent. Signal-to-noise ratios corresponding to 50% correct performance were measured. Results indicate that a pronounced Spanish accent limits a listener's ability to take advantage of cues to speech segregation and that a difference in accentedness between the target talker and background masker may be a useful cue for speech segregation. Older hearing-impaired listeners performed poorly in all conditions with the accented talkers.

Figures

Figure 1

Mean pure-tone thresholds (in dB hearing level, re: ANSI, 2010) of the three listener groups.

Figure 2

Mean speech recognition performance in noise (SNRs) by three listeners groups in six background noise conditions (NF = native female talkers, NM = native male talkers, NFM = native female + native male talkers, NNM = non-native male talkers, N + NNM = native + non-native male talkers, SMN = speech-modulated noise) for three talkers varying in accent (panel a = no accent, panel b = mild accent, panel c = moderate accent).

Figure 3

Mean speech recognition performance in noise (SNRs) of three listener groups for three talkers varying in accent (data collapsed across noise conditions). Brackets indicate significant talker effects within each group (p < 0.01).

Figure 4

Mean speech recognition performance in noise (SNRs) in the six noise conditions for three talkers (data collapsed across listener group).

Figure 5

Individual and mean group SNR data for the unaccented and moderately accented talkers in the native male (NM) and non-native male (NNM) background talker conditions.