A Multichannel Flexible Interface for Environmental-Robust Laryngeal Signal Decoding.

Area of Science:

• Wearable technology
• Biomedical engineering
• Artificial intelligence

Background:

• Conventional voice systems struggle in noisy or speech-impaired environments.
• Decoding subtle physical cues during speech is crucial for robust interaction.

Purpose of the Study:

• To develop a wearable, flexible sensor for decoding laryngeal and submandibular motion.
• To achieve high-accuracy speech recognition in complex acoustic and physical conditions.

Main Methods:

• Integrated a piezoresistive sensing layer (PDMS/PPy) on a flexible substrate (PI) with a multichannel electrode array.
• Developed a hardware platform for data acquisition, wireless transmission, and on-device processing.
• Employed a modified Audio Spectrogram Transformer (AST) with multichannel fusion for semantic recognition.

Main Results:

• Achieved 99.6% and 96.4% classification accuracy on structured datasets.
• Demonstrated robust performance under motion, facial expressions, and background noise.
• Validated the system's generalizability, semantic clarity, and stability in real-world scenarios.

Conclusions:

• The developed system significantly advances speech recognition capabilities in challenging environments.
• Offers a scalable solution for assistive communication, wearable AI, and silent interaction.
• Unifies soft materials engineering, flexible circuits, and multimodal deep learning for next-generation interfaces.