Early Autism Detection Through Vocal Pattern Analysis: A Machine Learning Approach Using Infant Cry Acoustics

Posted: Oct 16, 2025

• Wei Zhang
• Maria Rodriguez
• Kenji Tanaka
• Fatima Al-Jaber

Abstract

This study investigates the efficacy of machine learning algorithms in detecting early signs of autism spectrum disorder (ASD) through acoustic analysis of infant vocalizations. We collected and analyzed 2,500 audio samples from infants aged 6-18 months, including 1,200 samples from typically developing infants and 1,300 from infants later diagnosed with ASD. Using feature extraction techniques including Mel-frequency cepstral coefficients (MFCCs), pitch contours, and spectral properties, we trained and evaluated multiple classification models. Our results demonstrate that support vector machines achieved 87.3% accuracy in distinguishing ASD-related vocal patterns, with random forests and neural networks performing at 84.1% and 89.2% accuracy respectively. The findings suggest that automated vocal analysis could serve as a valuable screening tool for early ASD detection, potentially enabling earlier intervention and improved developmental outcomes.