Education
- Overall GPA 4.14 / 4.30, Summa Cum Laude
- Ranked 9th out of 434 students, 3rd within EE department of 116
Experiences
- Optimized speech generative models for client-side deployment
- Leveraging large-scale data analysis and teacher models to maximize the quality while compressing its size
- Researched and developed evaluation metrics and pipelines for voice generation models Paper
- Implemented an evaluation pipeline of voice generation models based on previous research
- The metrics have been actively used subsequently and contributions are referenced in the acknowledgments section of the published paper
- Led the development and deployment of an advanced E2E ASR system, achieving significant benchmarks in performance and scalability Model Specs
- Engineered a speech recognition engine to meet high performance standards in low-resource settings, integrating over 100M parameters and 20k hours of speech data
- Enhanced model robustness and effectiveness using unsupervised training techniques, achieving over 500 RTFx on AWS EC2 g4dnxlarge instances and reducing latency to 200ms
- Designed adaptable system architecture using SoTA techniques (e.g., ALiBi), and developed customized layers with CUDA and pybind11 for domain-specific B2B applications
- Applied the system in practical B2B scenarios and the VITO application, demonstrating its industry relevance Demo
- Developed a neural network language model for Hybrid ASR systems, employing neural network acceleration and knowledge distillation to enhance low-latency streaming ASR Related Article
Honors and Awards
KAIST Summa Cum Laude
2024/02
KAIST
Krafton AI Fellowship
2023/02
Evaluated based on AI knowledge, Awarded 10 million KRW, More Info
Krafton Inc.
KFAS Undergraduate Fellowship
2019/02
Korea Foundation For Advanced Studies
KAIST Presidential Fellowship
2017/02
Awarded to Science High School Valedictorian students, More Info
KAIST
KAIST Dean's List
2017/09 - 2020/09
Top 3% GPA for 4 out of 8 semesters
KAIST
OSS Contributions
SpeechBrain
- Contributor and maintainer of E2E ASR recipe
Skills & Proficiency
Programming Languages
Python, C++, C, Shell Script, MySQLLibraries & Toolkits
Pytorch, Fairseq, Kaldi, Triton Inference Server, Onnx, TensorRT, CMakeSoftware
Linux, Git, Docker, Docker ComposeLanguages
English
- TOEFL 104/120, TOEIC 990/990, TEPS 525/600 (Top 3.7%)
Others
Blog Articles
Article on the evolution from Hybrid to E2E ASR