▲ Park Ho-sung (supervisor: Professor Kim Ji-hwan) and Son Hyun-soo (supervisor: Professor Kim Ji-hwan)

Park Ho-sung and Son Hyun-soo (both supervised by Professor Kim Ji-hwan of the Department of Computer Science & Engineering), students at the University, received the Google Cloud Award at the 2nd Future Finance AI Challenge hosted by KB Financial Group and the Financial Supervisory Service. The competition was created to gather ideas for the future of finance using AI technology and to raise outstanding students' interest in the financial sector. A distinctive feature was the addition of a category using KB-ALBERT — the financial sector's first Korean natural language learning model, developed jointly by KB Kookmin Bank and Google Cloud Korea on the basis of the latest deep learning model ALBERT.

Applications were accepted for a month from August from undergraduates, graduate students and job seekers at major universities nationwide, and final rankings were decided through competitive presentations at the final round held in Lee Sam-bong Hall at Ewha Womans University on 22 September. One Grand Prize team, two Top Excellence teams and two Excellence teams were selected. All winners receive a share of 12.8 million won in prize money and exemption from the document and written test stages when applying to KB Kookmin Bank's open recruitment.

Park Ho-sung of the Sogang University team, which received an Excellence Award, explained what prompted them to enter: "We wanted to combine the various speech recognizers we are researching in the Auditory Intelligence Laboratory with the KB-ALBERT model we were given, and build a service that could actually be used."

Under the theme 'Hi, KB — an AI Assistant for KB Staff and Customers', the Sogang University team proposed an in-house AI speaker capable of handling customers' sensitive information in a contactless environment. They implemented and demonstrated on site keyword spotting to detect wake words for calling the AI speaker such as 'Hi, KB' and 'OK, Google'; speech recognition (STT) that takes the user's utterance and converts it to text for input to the dialogue model; a QA domain-based dialogue model that takes the user's request and provides an answer grounded in prior knowledge; and speech synthesis (TTS) that renders the dialogue model's output sentence in an AI voice for the user. The system promises improved customer satisfaction with contactless services, greater efficiency for bank staff, and an early lead in the market for financial transactions using AI speakers. The judges commented that the team's "demonstration, developed in-house so as to be genuinely usable rather than stopping at a plan, was impressive." The demonstration can be seen on YouTube (https://youtu.be/sz0wuLQ2WGk).

Son Hyun-soo said: "The preparation period was only about a week, so we could not be sure of a good result — it feels rewarding to have won," adding: "Through this experience I will work still harder to become a leading specialist in speech recognition."