[Courtesy of ETRI]
Voice recognition technology led by global tech companies such as Google, Microsoft, and Amazon, is widely used. Smart technology is mostly used in smartphones and AI-based voice assistant devices to understand voice commands based on a database of different languages.
Because each person has his or her unique choice of words and speaks in different styles and speeds, some voice recognition AIs are not very accurate. According to a consumer survey conducted in 2020 by Statistica, a global data analyst company, showed that 73 percent of consumers cited accuracy as the main factor that slows down the widespread adoption of voice recognition technology.
The Electronics and Telecommunications Research Institute (ETRI) said that its research team developed conversational AI technology capable of understanding 25 languages and converting conversations into text. The institute said that ETRI's AI is better at understanding Korean than global voice recognition technology leaders like Google.
Researchers said that the newly-developed AI uses various techniques such as self-guided learning, pseudo-label application, high-capacity multi-lingual learning model, and text-to-speech, to solve problems in processing different languages. A real-time streaming deduction model was developed to accelerate the processing speed.
ETRI will increase the number of supported languages to 30 by the end of 2022. "This development holds a great meaning as we were able to create voice recognition technology with a similar performance compared to global leading companies' AIs," ETRI researcher Kim Sang-hoon said in a statement on November 3.
reporter
Park Sae-jin
swatchsjp@ajunews.com
