A Novice Approach to Efficient Lip Reading

Vikas Kamra, Yash Deo, Dhruv Kalra, Kritika Singh, Shubham Kulshrestha · 2024

In an interconnected world, effective communication is crucial, especially for individuals with hearing impairments relying on lipreading. Artificial Intelligence and Machine Learning have revolutionized the traditional two-step process of manually ensuring clear lip visibility and understanding spoken words. This paper delves into the cutting-edge LipNet, the inaugural sentence-level lipreading model designed for end-to-end processing. In contrast to earlier models confined to word classification, LipNet achieves an impressive 95.2% accuracy in predicting sentences at the GRID corpus level. This achievement exceeds the performance of human lipreaders and surpasses the previous state-of-the-art word-level accuracy of 86.4%, marking a substantial advancement in lipreading technology.

Read the paper · More papers on PaperTik