我所论文：Throat map of speech recognition achieved by flexible ultrasensitive carbon array sensors with deep learning，在Top期刊Carbon（IF：11.6）上发表。

Throat map of speech recognition achieved by flexible ultrasensitive carbon array sensors with deep learning

Carbon 244 (2025) 120720

Peidong Xue, Kai Wang, Conghui Ma, Cheng Chen, Meijie Yin, Dongfeng Diao*

Laryngeal movements during speech provide a promising pathway for information transmission, yet the complex dynamics of vocal cord and surrounding muscles generate intricate signals across the throat, posing challenges on optimal sensor placement for signal obtaining. Here, we propose a “throat map” that leverages flexible carbon array sensor and deep-learning-based signal processing method to identify optimal sensor placement coordinates for high-accuracy speech recog­ nition. For the sensor fabrication, graphene nanocrystalline carbon films were deposited by electron cyclotron resonance as sensing units, then transferred onto PDMS substrates and integrated with stretchable circuits. The sensor consists of 16 sensing units within a 2 × 2 cm² area, achieving an ultrahigh gauge factor (>1000) and a frequency response limit of 10000 Hz. For signal processing, we originally proposed a signal-position classification method using a convolutional neural network algorithm. This approach visualizes the laryngeal prominence-centered coordinate information and constructs the throat map, enabling the identification of the position that generates the most distinctive and consistent signal. Leveraging the throat map, selected sensing units achieved >96 % accuracy in classifying 14 phonemes via a deep learning model. This throat map can serve as a guideline for sensor placement in speech and throat language recognition applications.