8th Speech in Noise Workshop, 7-8 January 2016, Groningen

Perception of American English consonants by Japanese listeners in background noise, reverberation, and background noise + reverberation

Hinako Masuda(a)
Waseda University, Tokyo, JP

(a) Presenting
(b) Attending

Numerous studies have shown that both native and L2 listeners’ phoneme identification are negatively affected by background noise and reverberation, and that the challenge is larger for L2 listeners. However, past research have often discussed the various listening environments separately. Although direct comparison between background noise and reverberation is difficult, it is important to examine the similarities and differences of environmental impact on L2 speech perception. The present study aims to overlook L2 speech perception in various listening environments (quiet, background noise, reverberation, and background noise + reverberation), and to reexamine the relationship between L2 speech perception in adverse environments and L2 proficiency. In particular, we focus on Japanese listeners' identification and confusion patterns of American English consonants in background noise, reverberation, and background noise+ reverberation. Twenty-three Japanese and 12 American English listeners participated in Experiment 1 (published), and 22 Japanese and 24 American English listeners participated in Experiment 2 (new experiment). Listeners were presented with 23 English consonants in intervocalic context under quiet, background noise, reverberation, and background noise + reverberation. Experiment 1 consisted of four listening environments; 1) quiet, 2) SNR = 10 dB, 3) SNR = 5 dB, and 4) SNR = 0 dB. Experiment 2 consisted of five listening environments; 1) quiet, 2) RT = 0.78 s, 3) RT = 1.12 s, 4) RT = 1.43 s, and 5) SNR = 10 dB + RT = 0.78 s. Correct identification rates and confusion patterns were analyzed against Japanese listeners' TOEIC(R) scores and LOR. Mean identification rates pooled across 23 consonants showed a significant difference between native and non-native listeners in Experiment 2 (p = 0.03) but not in Experiment 1 (p = 0.58). Correlation coefficient r showed that TOEIC(R) scores and listening environments had marginal correlations in Experiment 2, suggesting higher TOEIC(R) scores result in better identification in reverberation. Experiment 1 showed the opposite case, where correlation became weaker as listening environment became more adverse, suggesting even high TOEIC(R) score holders suffer identifying consonants with increased background noise. Weak to moderate correlation (r = 0.31 ~ 0.50) was observed between LOR and identification rates only in Experiment 2, with the lowest correlation in quiet environment, suggesting that experience of residing overseas may be beneficial for L2 perception in reverberation but not necessarily so in background noise. Further analysis is performed on the confusion patterns of higher and lower Japanese groups and compared against native listeners.

Last modified 2016-05-12 14:22:09