Lightweight and Robust ASR
Efficient, robust speech recognition with Zipformer and Conformer, from data curation and training to quantization and knowledge distillation.
Speech & Audio-Language Intelligence
Jongha Kim
M.S. Student at Sogang University
I am an M.S. student in Artificial Intelligence at Sogang University, advised by Prof. Hyung-Min Park in the Intelligent Information Processing Lab (IIP).
My research focuses on lightweight and robust automatic speech recognition (ASR) that works in real-world conditions, and on connecting ASR with language models.
I am also interested in large audio-language models (LALMs) that understand spatial and acoustic information beyond words.
Research interests
From recognition to understandingSpatial Audio-Language Models
Modeling spatial information from multichannel microphone inputs for audio-language understanding.
Acoustic Understanding and Reasoning
Connecting speech recognition and acoustic evidence with language models for spoken question answering and audio reasoning.
Selected research
All projects
Medical Speech Recognition and Spoken QA
Adapted Voxtral Mini 3B for Korean medical speech recognition and spoken question answering with 4-bit QLoRA and two-stage text/audio fine-tuning.
Medical ASR CER 19.14% → 9.09%; spoken QA 2.07 → 4.01/5, judged by GPT-OSS. Both comparisons use the Voxtral 3B 4-bit baseline in the project evaluation.
Intelligent Information Processing Lab (IIP), Sogang University
Audio-to-Audio Speech Unit Translation (Korean ↔ English)
Exploring Korean–English speech-to-speech translation through discrete speech units, without an intermediate text representation at inference time.
Generated English speech scored BLEU 42.8 / COMET 0.1009 in the project evaluation. Output was intelligible but had unstable pitch and background noise.
ECE Capstone Design, Inha University
Publications & reports
Jongha Kim, Leehyeon Song, Hyung-Min Park. “Task-Leaf Routed MiMo-Audio for DCASE 2026 Task 5.” DCASE 2026 Challenge, technical report, Jun. 2026. Official evaluation accuracy: 49.13% (Kim_SGU_task5_4). Report (PDF) · Challenge results
Jang Ji-hye, Lee Young-jun, Heo Ji-won, Kim Jong-ha. “Development of an ROS-based Environmental Perception and Decision-making System for Indoor Autonomous Mobile Robots.” Korean Society of Automotive Engineers (KSAE), Jeju, Korea, Oct. 2022. (Poster)
Experience
Full experienceGraduate Researcher
Mar. 2026 – PresentIntelligent Information Processing Lab (IIP), Sogang University
Research Intern
Jul. 2025 – Feb. 2026Intelligent Information Processing Lab (IIP), Sogang University · Presentation (PDF)
ML Scientist / Embedded Engineer
Mar. 2025 – Jun. 2025R&D Department, STS Engineering
Research Intern
Jan. 2024 – Dec. 2024Machine Intelligence Lab, Inha University
Research Intern
Jul. 2022 – Dec. 2022Intelligence Embedding System Lab (IESL), Inha University
Education
Sogang University, Seoul, Korea
Mar. 2026 – PresentM.S. Student, Department of Artificial Intelligence
Intelligent Information Processing Lab (IIP) · Advisor: Prof. Hyung-Min Park
Inha University, Incheon, Korea
Feb. 2019 – Aug. 2025B.S. in Electronic Engineering
GPA 3.98 / 4.5