Abstract
Acoustic speech recognition, as a technique to decode text from a speech, receives a great success in recent years. The trained model of Ping An Technology (ShenZhen) Co., Ltd results in a word error rate (WER) of 8.4%, which shows competitive performance among popular business products. However, an assumption of the achievement is the quiet environment of the speech. In a noisy environment, the accuracy will decrease 10%–20%. For the improvement in such environment, a multi-modal biometric system integrating acoustic speech-recognition with sentence level lip-reading is designed. In several noisy situations, the 5.7% averaged word error rate (WER) of the results of our integrated system indicates a significant improvement to the pure acoustic speech-recognition system.
Author supplied keywords
Cite
CITATION STYLE
Wang, J., Wang, Y., Liu, A., & Xiao, J. (2017). Assistance of Speech Recognition in Noisy Environment with Sentence Level Lip-Reading. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 10568 LNCS, pp. 593–601). Springer Verlag. https://doi.org/10.1007/978-3-319-69923-3_64
Register to see more suggestions
Mendeley helps you to discover research relevant for your work.
