Mandarin Speech Recognition in Babble for Chinese- and Korean-Native Listeners

This study aimed to investigate the effects of noise on Mandarin sentence perception for listeners with different language backgrounds. The stimuli employed consisted of 112 Mandarin sentences "ready [call sign] go to [color] [number]" from the coordinate response measure (CRM) corpus, for example, "ready Li Ting go to green seven now." Speech reception thresholds (SRTs) in identifying these sentences were measured among Chinese-native (CN) listeners, and Korean-native listeners with high (KNH) and medium (KNM) Mandarin Chinese proficiency. The target speech signals were presented in five types of babble noises: one-talker babble of a different gender (1Diff), one-talker babble of the same gender (1Same), two-talker babble (2TB), four-talker babble (4TB), eight-talker babble (8TB). Overall, 4TB noise brought the most difficulty for all three groups of listeners. When the number of talkers in babble was less than 4, the more talkers in babble, the higher SRT. Compared to the 4TB, the SRT was only slightly reduced at the 8TB. These results were consistent with the findings of English CRM versions. Moreover, non-native listeners underperformed native listeners in the noise of 1Diff, 1Same, and 2TB. The gap between native and non-native listeners became smaller as the number of talker increased up to four and then almost disappeared at 8TB. These results imply that when the number of talkers was small like 1-2 talkers, informational masking of babble was remarkable, likely resulting in more challenges to non-native speakers than native speakers. As the number of talkers in the babble increased up to 4-8, the informational masking of babble was reduced to an ignorable level, such that native and non-native speakers had similar performance.