현재 Spring + PS 준비중 => 앱 개발중
Overview: The MADUV Challenge 2025 was part of INTERSPEECH 2025, inviting researchers worldwide to develop models that classify mice as either wild-type or ASD models based on high-frequency ultrasound vocalizations. Advanced signal processing techniques were used to handle complex audio data in this globally recognized event.
My Contribution:
- Swin Transformer Implementation: Explored multiple Swin Transformer variants by reviewing recent research papers, implementing the approaches, and conducting thorough experiments.
- Mel-Spectrogram Conversion Pipeline: Developed an end-to-end pipeline to convert raw ultrasound vocalizations into mel-spectrograms for robust feature extraction.
Participated in a Statistics Competition organized by the Department of Statistics at Konkuk University.
EmoChat – Real-time Emotion & Stress Voice Chat App ( Personal Project : FE + Baas + AI research + deploy )
Overview: An end-to-end voice-based chat application that captures speech, transcribes it using OpenAI Whisper, predicts emotional stress levels via Hugging Face audio model, and reflects the result in real-time chat UI (e.g., warning modal, balloon color shift).
Highlights:
- 📱 Built with React Native (Expo), customized GiftedChat UI
- 🧠 AI integration: Whisper STT + Hugging Face Inference API
- 🗂 Realtime BaaS: Firebase Authentication, Firestore, Cloud Storage
- 📊 Integrated lightweight voice-only stress recognition model via knowledge distillation
📄 View EmoChat Project Summary PDF
(Personal Project · FE + BE + AI Research + DevOps · Journal/Conference paper under preparation)
Overview Browser-based chatbot that listens to the user’s voice, transcribes it (Whisper), estimates emotion (7-class) and stress level (2-class StudentNet), then generates a persona-aware GPT-4 reply—everything rendered live in Streamlit with dynamic badges, gauges, and RAG-augmented context.
Highlights
| 영역 | 핵심 포인트 |
|---|---|
| 🖥 Frontend | Streamlit + Tailwind-style CSS · Mic/WebRTC · custom badges & stress gauge |
| 🧠 AI Stack | Whisper STT · DistilRoBERTa emotion clf · StudentNet stress clf (self-trained, knowledge-distilled, F1 +17 pp) |
| 🗄 Context/RAG | GPT-4 + Pinecone Vector DB → persona-specific, context-aware replies |
| 🧰 Backend | Python 3.10, modular ChatbotService (LLM · emotion · stress · RAG) |
| 📑 Research Output | 예정 |
Links • Demo Slides (MiriCanvas): https://www.miricanvas.com/v/14sl4ew • GitHub Repo: https://github.com/forwarder1121/SER-CHATBOT-PROJECT
-
Languages:
- C, C++
- JavaScript (Node.js, React)
- Python (Django)
- Java, R
-
Databases:
- MongoDB, MySQL
- Actively solving problems on Baekjoon Online Judge
- Data Structures & Algorithms (Baekjoon)
- React.JS & Node.JS – In-depth study of modern JavaScript frameworks
-
Start Date: March 28, 2024
-
Resources:
- Deep Learning from Scratch
- CS231n
