Development and analysis of analog-digital neural net for speech stress detection

Autonomous SSDs (Speech Stress Detector) are widely used at present. Known electronic realizations suffer from low noise immunity, their methods of stress detection require further development and verification. State-of-the-art integrated technologies permit to superpose digital reprogrammable systems with analog neuron like elements and receive new quality of data processing in SSD. Proposed real-time neural network is characterized by rational representation and distribution of main SSD functions in neural net in analog and digital forms. Single layer structure of ADNN (Anolog-Digital Neural Net) supports self-organization process in SSD during speech analysis, new features extraction and classification. Developed by the authors ADNN SSD prototype has shown higher noise immunity level in comparison with conventional realizations, can be treated as promising model for instrumental base development on the one hand and hardware real time realization in the future embeddable systems on the other.