Chen Fei, Guan Tian, Wong Lena L N
Annu Int Conf IEEE Eng Med Biol Soc. 2013;2013:4199-202. doi: 10.1109/EMBC.2013.6610471.
Temporal fine structure (TFS) carries important information for the speech perception of hearing-impaired listeners and for the design of novel prosthetic hearing devices. This study assessed the performance of present intelligibility indices for predicting the intelligibility of speech containing different amount of TFS information. Speech intelligibility data was collected from vocoded and wideband Mandarin sentences containing little/partial and intact TFS information, respectively, and was then subjected to the correlation analysis with existing intelligibility indices. It was found that, though performing well in predicting the intelligibility of vocoded or wideband speech separately, present intelligibility indices were not highly correlated with the intelligibility scores when a general function was used to map all intelligibility measures to intelligibility scores. Analysis further showed that the intelligibility prediction power could be significantly improved when multiple condition-dependent functions were used for mapping intelligibility measures to intelligibility scores.
时间精细结构（TFS）为听力受损听众的言语感知以及新型人工听觉装置的设计承载着重要信息。本研究评估了当前可懂度指标在预测包含不同数量TFS信息的言语可懂度方面的性能。分别从包含少量/部分和完整TFS信息的声码化和宽带普通话句子中收集言语可懂度数据，然后将其与现有的可懂度指标进行相关性分析。结果发现，尽管当前的可懂度指标在分别预测声码化或宽带言语的可懂度方面表现良好，但当使用一个通用函数将所有可懂度度量映射为可懂度分数时，这些指标与可懂度分数的相关性并不高。进一步分析表明，当使用多个依赖条件的函数将可懂度度量映射为可懂度分数时，可懂度预测能力可得到显著提高。
