Adaptive Dual-Channel Pulse-Coupled Neural Network with S-Transforms for Speech Source Localization Using Binaural and Monaural Microphone Arrays in Hearing Aids
G. Ramani, D. Kavitha, G. Charlyn Pushpa Latha, M. Siva Ramkumar, Jayant P Giri, Khaled E. Al-Qawasmi · 2025
The S-transforms-based Adaptive Dual-Channel Pulse-Coupled Neural Network with Skill Optimization Algorithm (SADCPCNNet-SOA) is a novel framework that enables Hearing Aids (HAs) equipped with binaural and monaural microphone arrays to localize speech sources in environments that are noisy and reverberant. The framework pre-processes microphone channels using multiple discrete orthonormal S-transforms (MDOST), simulating the human auditory system. These subband signals are then used by the Adaptive Dual-Channel Pulse-Coupled Neural Network with Skill Optimization Algorithm (ADCPCNNet-SOA) to reliably identify the voice source. In binaural setups, the model performs well with 97% accuracy, 96% precision, and 95% recall. In monaural configurations, it surpasses previous methods with up to 98% accuracy and 97% recall.