Skip to content
View leeytkfng's full-sized avatar
🎯
Focusing
🎯
Focusing
2 followers · 0 following

Achievements

Achievement: YOLOAchievement: Pull Shark

Achievements

Achievement: YOLOAchievement: Pull Shark

Block or report leeytkfng

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
leeytkfng/README.md

이용수 (Yong-Su Lee)

AI Researcher in Audio Deepfake Detection
μ˜€λ””μ˜€ λ”₯페이크 탐지 Β· μŒμ› 뢄리 Β· μ„€λͺ… κ°€λŠ₯ν•œ AI

πŸ”¬ Research

ν˜„μž¬ 볡합 μ˜€λ””μ˜€ μœ„λ³€μ‘° 탐지 연ꡬλ₯Ό 진행 μ€‘μž…λ‹ˆλ‹€.
κΈ°μ‘΄ λͺ¨λΈμ΄ "μ§„μ§œ/κ°€μ§œ" 이진 νŒλ³„μ— κ·ΈμΉ˜λŠ” 것과 달리, μŒμ„±κ³Ό ν™˜κ²½μŒμ„ λΆ„λ¦¬ν•˜μ—¬ "μ–΄λŠ 채널이 κ°€μ§œμΈκ°€" λ₯Ό 5개 클래슀둜 μ„€λͺ…ν•˜λŠ” ν”„λ ˆμž„μ›Œν¬λ₯Ό μ œμ•ˆν•©λ‹ˆλ‹€.

관심 연ꡬ λΆ„μ•Ό

  • μ˜€λ””μ˜€ λ”₯페이크 탐지 (Audio Anti-Spoofing)
  • μŒμ› 뢄리 (Blind Source Separation)
  • μ„€λͺ… κ°€λŠ₯ν•œ AI (Explainable AI, XAI)
  • μŒμ„±/μ˜€λ””μ˜€ μ‹ ν˜Έ 처리

πŸš€ Featured Research Project

S-CAD: Spatial-Consistency-Aware Audio Detection

μŒμ› 뢄리 + 물리 음ν–₯ ν”Όμ²˜ 기반 5-class 볡합 μ˜€λ””μ˜€ μœ„λ³€μ‘° 탐지 μ‹œμŠ€ν…œ

ν•­λͺ© λ‚΄μš©
데이터셋 CompSpoofV2 (24,864 samples, 5-class)
핡심 λͺ¨λΈ Conv-TasNet (μŒμ› 뢄리) + LightGBM (19-dim 물리 ν”Όμ²˜)
Accuracy 69.8% Β· Macro-F1 0.6564
FAKE Recall 56.4% (SuDORM-RF λŒ€λΉ„ +41%p κ°œμ„ )
λ…Όλ¬Έ μ •λ³΄μ²˜λ¦¬ν•™νšŒ λ…Όλ¬Έ μž‘μ„± 쀑 (2026)

핡심 κΈ°μ—¬

  • μ‚¬μ „ν•™μŠ΅ 없이 물리 음ν–₯ μ§€ν‘œ(RT60, Noise Floor, MSC, XCorr)만으둜 경쟁λ ₯ μžˆλŠ” 탐지 μ„±λŠ₯ 달성
  • SI-SNR ↔ FAKE recall λΉ„μ„ ν˜• 상관관계 μ •λŸ‰ 싀증
  • LightGBM Feature Importanceλ₯Ό ν†΅ν•œ νŒμ • κ·Όκ±° μ‹œκ°ν™”
[μž…λ ₯ μ˜€λ””μ˜€]
     β”‚
[LCNN-SE Gatekeeper]  β†’  REAL μ‘°κΈ° νŒμ •
     β”‚
[Conv-TasNet μŒμ› 뢄리]
     β”œβ”€β”€ πŸ—£οΈ Speech Stream β†’ WavLM μœ„μ‘° 점수
     └── 🌿 Env Stream   β†’ LCNN-SE μœ„μ‘° 점수
     β”‚
[물리 음ν–₯ 뢄석: RT60 Β· Noise Floor Β· MSC Β· XCorr]
     β”‚
[LightGBM 5-class] β†’ REAL / GENUINE / SPOOF_SPEECH / SPOOF_ENV / FAKE

πŸ›  Tech Stack

AI / ML
Python PyTorch LightGBM scikit-learn HuggingFace

Audio / Signal Processing
librosa torchaudio SpeechBrain

Backend (κΈ°μ‘΄ κ²½ν—˜)
Java Spring Boot PostgreSQL Docker

πŸ“„ Publication

μƒνƒœ 제λͺ© ν•™μˆ λŒ€νšŒ
μž‘μ„± 쀑 S-CAD: μŒμ› 뢄리 기반 볡합 μ˜€λ””μ˜€ μœ„λ³€μ‘° 5-class 탐지 μ •λ³΄μ²˜λ¦¬ν•™νšŒ 2026

πŸ“« Contact

Top Langs

GitHub Streak

Popular repositories Loading

  1. TodoApi TodoApi Public

    Java

  2. MSA-Perpomance-Service MSA-Perpomance-Service Public

    user MicroService Architecture

    Java 1

  3. Msa-Board-Service Msa-Board-Service Public

    Java

  4. MSA-Flight-Service MSA-Flight-Service Public

    JavaScript

  5. leeytkfng leeytkfng Public

  6. Test-CI Test-CI Public

    JavaScript