FBSE-FTFCWT-based automated detection of specific language impairment in children using speech signals

Abstract

Specific language impairment (SLI) is a communication disorder in children that hampers their ability to speak or interpret spoken words. Early detection of SLI can be helpful in speech therapy for these children. This paper proposes an automated system for detection of SLI using the Fourier-Bessel series expansion-based flexible time-frequency coverage wavelet transform (FBSE-FTFCWT) and synchrosqueezed wavelet transform (SSWT)-based marginal spectral density (MSD). The FBSE-FTFCWT decomposes speech segments into subband signals. The MSD features are extracted from these subband signals. The studied temporal convolutional network is trained using MSD features. A speaker-independent study is performed on the database provided by laboratory of artificial neural network application (LANNA) in which 10% dataset is separated for testing purpose. The presented study includes analysis of the proposed framework on several subband signals. The results indicate that there is a significant amount of information present in high-frequency subband signal resulting in better classification of speech signals corresponding to healthy subjects and subjects with SLI. The proposed framework demonstrates better performance over existing state-of-the-art approaches, achieving accuracy of 99.86%, precision of 99.97%, recall of 99.83%, specificity of 99.92%, and F1-score of 99.90%. © 2026 Elsevier Ltd