Engineering case study / 2023
Echotion
Connecting speech, emotion classification, and expressive subtitles.
PM · Development · Two-person course project

The goal
Echotion explores how subtitles can communicate more than the words in speech. The course project connected speech recognition to emotion classification and a desktop interface that displays expressive subtitles.
My contribution
I worked as PM and developer in a two-person team. The implementation connected speech recognition, a KoBERT-based emotion classifier, and the subtitle interface.
The integration problem
Recognition, classification, and presentation are different stages. The interface needs the recognized text, while the visual expression also depends on the classifier's output. Bringing these stages into one application was the core implementation task.
The workflow moves from speech input to recognized text, then to emotion classification and subtitle presentation. The visible subtitle is where the outputs meet: the words communicate the content and the presentation adds an emotional cue.
Outcome and takeaway
The result was a desktop course-project prototype. The application brought speech input, classification, and subtitle rendering into one interface.
The takeaway is to treat the interface as part of the pipeline. A model output only becomes an interaction when it is translated into something a person can perceive and interpret.