Effects of training on the perception of talker information in degraded speech

Abstract

A fundamental property of human speech perception is its robustness in a wide range of listening conditions. Listeners are able to successfully recognize and understand speech under a wide range of adverse and challenging conditions, such as in noise, reverberation, or multi-talker babble, and with noise vocoding. Noise vocoding, a manipulation that reduces spectral resolution, has been shown to result in less accurate speech comprehension and also poorer perception of indexical information (e.g., talkers’ voices). However, speech comprehension quickly improves with exposure, as listeners learn to better extract information from the systematically degraded speech. The extent to which the learning of noise-vocoded speech transfers to other talkers or tasks is still largely unknown. The current study investigated whether training, consisting of transcribing sentences or identifying talkers’ voices from sentences, leads to improvements in the perception of talker information. Preliminary results suggest that both training tasks resulted in improved perception of talker information, as participants learned general information about the degradation. Larger improvements were observed after talker identification training, which focused attention on talker differences. Differences between training tasks, and implications for cochlear implant users, will be discussed. [Funding: VENI Grant (275.89.035) from the Netherlands Organization for Scientific Research (NWO).]