Flexible TENG Sensors and Deep Learning Approaches for Silent Speech Recognition

Abstract Silent speech recognition (SSR) holds significant promise for assisting individuals with vocal impairments and enabling secure communication in noise‐sensitive environments. In this context, triboelectric nanogenerators (TENGs) have attracted considerable attention as efficient, wearable energy harvesters with great potential for sensing applications. Their unique properties make TENGs well‐suited for applications in SSR, enabling noninvasive and sensitive signal acquisition that can enhance the performance of such systems. In this study, a novel deep learning approach for SSR is proposed, employing TENGs as self‐powered sensors to capture biomechanical signals from lip movements during silent speech articulation. For sensor architecture design, micropatterned polydimethylsiloxane (M‐PDMS) is fabricated via sandpaper imprinting. Experimental results demonstrated that the output performance of M‐PDMS is 3.5 times higher than that of pure PDMS, confirming its improved energy output performance. A one‐dimensional convolutional neural network (1D‐CNN) integrated with an attention mechanism is developed to enable efficient feature extraction and classification, achieving an average accuracy of 95% across ten distinct word categories. To visualize recognition results, a robotic hand is controlled via serial communication based on the classification output. The system demonstrates robust and data‐efficient, highlighting its potential as wearable and user‐friendly interface for voiceless human–machine interaction.