CAC: Content-Aware Captioning for Professional Online Lectures in Korean Language

ABSTRACT

The use of online video conferencing has skyrocketed due to the ongoing outbreak of COVID-19. Not only professionals and college students, but even the elders and preschoolers use video meeting programs. As such, and due to a variety of reasons such as lack of context or audio-incapable circumstances, the need for high quality closed-caption has also risen. Speech-to-Text (STT) technology has been leaping forward significantly, yet involving the defect of the somewhat low recognition rate of jargons and professional terminology, which can lead to critical misunderstandings for lectures or meetings dealing with field-specific contents. To address this issue, we propose an enhancement to the recognition of technical terms by introducing Contents-aware-captioning (CAC) algorithm which applies automatically extracted data from previous materials of the wehinar or terminology dataset according to the subject.