Principal Investigator: Chng Eng Siong (Assoc Prof , School of Computer Science and Engineering (SCSE), Assistant Chair (Graduate Studies))
Team: Dr. Xu Haihua (Senior Research Scientist B, Temasek Laboratories), Lim Zhi Hao (Project Officer, SCSE), Vu Thi Ly (Research Associate, SCSE) & Kyaw Zin Tun (Project Officer, SCSE)
Mentor: Kee Thian Seng (Director, K&L Engineering Services)
Customizable speech-to-text engine that can better recognize local Singlish
Modelling long natural conversations is now possible with the advancement of deep learning. Modern challenges presented by mic type, room acoustics, mic-distance-to-mouth or domain lingo and an all-for-one speech engine remain out of reach. A speech engine tuned for telephones may fail when used in meeting recordings. In order to achieve high levels of accuracy in the industry, customisation is still key. Firstly, using our custom Singlish database, we have developed a speech recognition engine that is truly for Singaporeans.
The localised speech engine consists of Singaporean slang, street names and lingo. Everything from a local chatbot to voice-based virtual assistant needs. Secondly, by leveraging on speech research and engineering expertise developed over the years, AI SpeechLab aims to streamline the previous cumbersome customisation process into a painless procedure. This quick and effortless customisation process means the SpeechLab can tap into the $400 million market of English speaking APAC population more efficiently.