Print the page
This project aims to develop a bilingual speech recognition system that can accurately recognize continuous sentence level Cantonese and English speech inputs from impaired adult speakers with varying forms of speech disorders.
The system targets both conversational speech that is essential for daily use, and more formal style speech. The recognition vocabulary covers over 80 thousand common and uncommon words. The average recognition error rates will be below half of those obtained using the Google speech recognition engine. This system can aid the communication of the elderly and disabled people with speech disorders, improve their quality of life, social inclusion and the cost-effectiveness of the care they require.
Uniqueness and Competitive Advantages:
- Large accuracy improvements over commercial speech recognition systems
- Robust performance across varying forms of speech dysarthria
- Fast user adaptation using minimum data
- Audio-visual disordered speech recognition
- Low published word error rate