Developmental Changes in Speech Frequency Weighting in Children

Estimates of speech intelligibility depend on the audibility of spectral bands in speech and the importance weight of each spectral band for speech recognition. In children, the transfer function that links estimates of speech intelligibility to recognition accuracy changes with age, indicating the need to incorporate developmental changes into clinical use of speech intelligibility metrics for children. Previous research has shown developmental shifts in how children use spectral cues to support speech recognition, but individual differences in band importance that form the basis for models of speech recognition have not been directly validated in children. The goal of the present study was to determine if children differ in band importance weights with age and, if such differences exist, determine how they relate to developmental changes in vocabulary and cognitive ability.

Children (ages 6 to 13 years) with normal hearing (N = 42) completed a filtered word recognition task and a battery of vocabulary and cognitive assessments. Monosyllabic words and nonword auditory stimuli were filtered into six frequency bands that were centered on octave frequencies between 250 and 8000 Hz, each one octave wide. Each stimulus contained a pseudorandom combination of three out of six frequency bands, which children listened to and verbally repeated. Recognition accuracy was regressed against the presence or absence of each band across trials in a generalized linear mixed effects model to measure the contribution of each frequency band to speech recognition on an individual basis. Children also completed the Peabody Picture Vocabulary Test, the Wechsler Abbreviated Scale of Intelligence vocabulary task, the Psychology Experiment Building Language flanker and switcher tasks, and the Wechsler Abbreviated Scale of Intelligence matrix reasoning task.

Changes in frequency importance weights with age were observed for bands centered on 2000 and 4000 Hz, with increasing weights for these bands in older participants. Frequency bands centered on 500, 1000, and 2000 Hz had higher weights for words than for nonwords. Interactions between age and stimulus type were not evident. None of the vocabulary and cognitive tasks predicted frequency importance weights after controlling for age. Flanker task performance was a significant predictor of recognition accuracy.

Frequency band importance weights undergo developmental changes in children. Models used to predict speech recognition in children may need to be updated to account for these developmental changes; otherwise, speech intelligibility estimates will likely overestimate speech recognition accuracy. Additional work is needed to determine whether the developmental trajectory of frequency importance weights is affected by hearing loss in children who are hard of hearing.