Perceptual cues of consonant sounds and impact of sensorineural hearing loss on speech perception

speech perception;hearing loss;consonant;perceptual cue

Li, Feipeng

关键词: speech perception;    hearing loss;    consonant;    perceptual cue;   

Others  :  https://www.ideals.illinois.edu/bitstream/handle/2142/14603/Li_Feipeng.pdf?sequence=2&isAllowed=y

美国|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

This research investigates the impact of various types of cochlear hearing loss and mask-ing noise on the perception of basic speech sounds based on the information of identified speech cues. A psychoacoustic method, named three-dimensional deep search (3DDS), is developed to identify the perceptual cues of consonant sounds in natural speech. Unlike the conventional method of synthetic speech, which requires a prior hypothesis about the acoustic cues to generate the speech stimuli, the 3DDS measures the contribution of each subcomponent to speech perception as a function of time, frequency and intensity, without making any tacit assumptions about the speech cues to be identified.Using the 3DDS, we discovered that natural speech often contains conflicting cues that are characteristic of confusable sounds. For instance, a normal /ka/, dominated by a mid-frequency burst at 1–2 kHz, may also have an inaudible /ta/ burst above 3 kHz that promotes the /ka/→/ta/ confusion under noisy environments. Removal of the /ka/ burst may turn the sound into a solid /ta/. More than a dozen hearing-impaired ears were tested on consonant identification in noise. While the deterioration in performance for flat mild-to-moderate hearing loss can be well predicted by the loss of audibility, subjects with other types of hearing loss often show patterns of difficult sounds that can hardly be explained by the shift of hearing threshold. A subject with almost identical binaural hearing loss is nearly deaf to /ka/ in one ear due to a mid-frequency cochlear dead region. Among the 18 /ka/s produced by different talkers, the subject can only hear one /ka/ at an accuracy of 80% and three other /ka/s at 20–40%. Most /ka/s are highly confused with /ta/ because the subject is listening to the conflicting /ta/ burst in the high-frequency. The /ka/→/ta/ confusion is significantly reduced when the conflicting cue is removed. NAR-L improves the average score by 10%, but it may degrade a few consonants under certain circumstances.