Abstract: Lip language is an effective form of verbal communication for patients with vocal cord injuries, laryngeal and tongue injuries, and hearing loss. Lip-speaking signals are generated by lip and facial muscle movements, which contain a large amount of speech information. The extraction and recognition of lip-speaking signals can be achieved by capturing the muscle movements through flexible pressure sensors, providing a more natural and convenient way of accessible communication for patients with listening and speaking dysfunction. In this study, a piezoresistive flexible pressure sensor was prepared using a two-dimensional material, MXene, and a highly conductive polymer, poly(3, 4-ethylenedioxythiophene):poly(styrenesulfonate) (PEDOT:PSS), as composites, and a stretchable and microstructured ecoflex as a flexible substrate. The piezoresistive sensor demonstrates a high sensitivity of 42.31 kPa⁻¹ and fast response (<150 ms) in the pressure range of 0-2.5 kPa, and shows high stability in10000 compression-release cycles. The flexible piezoresistive sensor was attached to the corners of the mouth and captured the muscle movements of the lips, which was combined with a convolutional neural network algorithm to train and test the signals of English words in the Chinese zodiac, with an average accuracy of up to 90.18%. This work increases the versatility of lip recognition systems and lays an important foundation for the direct conversion of lip movement signals into speech or text.