AI speaker is now everywhere. For example, if you are in a KIA car, “Hi, Kia!’ should boot the car’s system. When most of the technical studies have focused on accuracy, we were curious about the emotional state contained in the voice. Led by Taesu, a Ph.D. candidate constructed a voice database recorded in various emotional states. Wake-Up Word (WuW) is a short voice recording of about 0.6 seconds. By extracting the acoustic properties of the voice data and incorporating machine learning analysis techniques, the machine-learning-based algorithm predicts the speaker’s mood at a (reasonably) satisfactory level of accuracy.
The study is published as a research article entitled, “Hi, KIA! Classifying Emotional States from Wake-up Words Using Machine Learning [Hi, KIA! 기계 학습을 이용한 기동어 기반 감성 분류]” in Journal of Korean Society for Emotion and Sensitivity[감성과학], 24(1), 91-104. This study is the spin-off research from the industry-academia project with KIA Design Center and Color Lab in 2020.