Sustainable Live Sound Monitoring and Classification System Enabled by a Triboelectric Nanogenerator and Machine Learning Techniques

Abstract

The growing demand for sustainable, real-time audio processing drives innovations in sound classification and energy harvesting. Traditional sound monitoring systems often struggle with scalability, energy efficiency, and adaptability, particularly in remote or resource-limited environments. The expansion of IoT applications intensifies power demands in widely distributed wireless sensor networks, highlighting the need for sustainable solutions. Moreover, the volume of data generated by these sensors frequently exceeds the capacity for efficient human analysis, necessitating the integration of machine learning and deep learning techniques. These methods must be optimized for fine-tuning with minimal data from new sensors, enabling efficient and accurate sound classification without extensive retraining. This paper presents a Triboelectric Nanogenerator (TENG)-based microphone that addresses energy consumption and data processing challenges by integrating advanced materials with sound classification systems. The proposed device uses polyimine/graphite polypropylene (PI/GP) coated paper to capture sound and harvest energy from ambient noise. It delivers an output power of 25.67 μW at 94 dB, powering a wireless transmission circuit while achieving high acoustic sensitivity and a frequency response of up to 20 kHz. Performance evaluations show 92.7% classification accuracy in simulated live environments and a processing time of 0.342 s for 5-s audio clips using the MobileNet V1 model. Pre-trained models fine-tuned with minimal data from the TENG microphone enable efficient sound classification without extensive retraining. This innovation offers a sustainable alternative to conventional microphones, supporting self-powered, real-time monitoring systems with wireless data transmission and energy storage capabilities.