Human infants demonstrate great knowledge about the sounds of their native language at birth, even though the uterine environment restricts their auditory perception to low-frequency ranges. This study explores the possibility that intrauterine low-pass filtering allows infants to extrapolate from prenatal impoverished input to postnatal speech sound knowledge. We trained neural network models in two stages, simulating prenatal and postnatal learning, and measured the impacts of natural low-frequency filtering, artificial high-frequency filtering, as well as full-frequency prenatal exposure, on speech sound learning. Three model architectures were utilized in the analysis: a long short-term memory-based neural network, a convolutional neural network, and a residual neural network. Results indicated that exposure to low-frequency sound input led to accelerated phonetic learning upon introduction of full-frequency sounds. In addition, the low-frequency filtering condition yielded better learning on phones compared to the high-frequency filtering condition during prenatal learning. These findings suggest the role of prenatal exposure to low-frequency sounds in enhancing infants' speech sound learning capabilities. (PsycInfo Database Record (c) 2026 APA, all rights reserved).
Zheng et al. (Thu,) studied this question.