Xuanjun Chen

Xuanjun Chen (陳炫均)

Ph.D. Candidate, NTU EECS · Taipei, Taiwan

I am a Ph.D. candidate at National Taiwan University, advised by Profs. Hung-yi Lee and Jyh-Shing Roger Jang. My research focuses on Speech & Audio Processing and Large Language Models, specifically audio under­standing, audio generation, agentic LLMs, and deepfakes. I have authored 15+ first/co-first papers in Speech/NLP venues (e.g., COLM, TIST, TASLP, ICASSP, INTERSPEECH, ASRU, SLT), and am a core con­trib­utor to DeSTA2.5-Audio (TASLP) and SUPERB bench­marks (ICLR, ACL). I hold a  Google PhD Fellowship (one of 13 in Machine Perception worldwide),  NVIDIA Academic Grant, and IEEE SPS IEEE SPS Scholarship, plus an IEEE ASRU Best Student Paper Nomination and three Best/Outstanding Paper Awards. I am actively looking for a 2027 research internship!

Feel free to reach out at d12942018 [at] ntu.edu.tw, or find him on LinkedIn, GitHub, and 𝕏.

Research Highlights

See all on Google Scholar or browse by topic.

Speech & Audio Language Models and GenerationSpeech & Audio LMs and Generation

Can LLM Agents Automate Reinforcement Learning for Text-to-Speech?

Xuanjun Chen, Zixiong Su, Hao Shi, Chang Zeng, Kai Li, Jyh-Shing Roger Jang, Hung-yi Lee

Technical Report, Oct. 2026

What Does LALM Post-training Learn? Rethinking Audio Understanding and Reasoning Capabilities

Hua-Hsuan Chen*, Xuanjun Chen*, Wei-Chung Lu, Chih-Kai Yang, Yu-Xiang Lin, Sung-Feng Huang, Haibin Wu, Hung-yi Lee

Technical Report, Oct. 2026

DeSTA2.5-Audio: Toward General-Purpose Large Audio Language Model with Self-Generated Cross-Modal Alignment

Ke-Han Lu, Zhehuai Chen, Szu-Wei Fu, ..., Xuanjun Chen, ..., Boris Ginsburg, Yu-Chiang Frank Wang, Hung-yi Lee

IEEE TASLP 2026JCR Q1, IF 5.173 Citations as of Oct. 2026

Towards audio language modeling – an overview

Haibin Wu, Xuanjun Chen, Yi-Cheng Lin, Kai-wei Chang, Ho-Lam Chung, Alexander Liu, Hung-yi Lee

Technical Report, Feb. 202488 Citations as of Oct. 2026

Retrieval and Agentic Large Language ModelsRetrieval & Agentic LLMs

MetaBench-Harness: Unlocking End-to-End Optimization of Benchmark Harnesses

Xuanjun Chen, Hua-Hsuan Chen, Wei-Chung Lu, Yinghao Ma, Jyh-Shing Roger Jang, Hung-yi Lee

Technical Report, Oct. 2026

Only Ask What You Don't Know: Grounded Delta Planning for Efficient Multi-step RAG

Wei-Chieh Chou*, Xuanjun Chen*, Jian-Ren Lin, Claire Lin, Hung-yi Lee, Jyh-Shing Roger Jang

COLM 2026Accept Rate 29.1%

CodaRAG: Connecting the Dots with Associativity Inspired by Complementary Learning

Cheng-Yen Li*, Xuanjun Chen*, Claire Lin, Wei-Yu Chen, Wenhua Nie, Hung-yi Lee, Jyh-Shing Roger Jang

ACM TIST 2026JCR Q1, IF 10.7

Audio Deepfake Detection

CodecFake+: Codec-Based Resynthesized Data as a Proxy for Detecting CodecFake Speech

Xuanjun Chen*, Jiawei Du*, Haibin Wu, Lin Zhang, I-Ming Lin, ..., Jyh-Shing Roger Jang, Hung-yi Lee

IEEE TASLP 2026JCR Q1, IF 5.1

Towards Generalized Source Tracing for Codec-Based Deepfake Speech

Xuanjun Chen*, I-Ming Lin*, Lin Zhang, Haibin Wu, Hung-yi Lee, Jyh-Shing Roger Jang

IEEE ASRU 2025Best Student Paper Nomination (Top 2.5%)

Singing Voice Graph Modeling for SingFake Detection

Xuanjun Chen, Haibin Wu, Jyh-Shing Roger Jang, Hung-yi Lee

INTERSPEECH 2024 (Oral)Lightning Talk @ SLT '2435 Citations as of Oct. 2026

Experience

  • : NVIDIA Academic Grant Program, Research Collaborator, Taipei, Taiwan
  • : National Taiwan University, Ph.D. in Communication EngineeringCommunication Eng. ·  GPA 4.3/4.3
  • : National Taiwan University, M.S. in Computer ScienceComputer Science ·  GPA 4.19/4.3
  • : Taiwan Tech, B.S. in Computer ScienceComputer Science ·  GPA 4.11/4.3

Open Source Contributions

Lead & Mentored Projects

  • : CodecFake+ (IEEE TASLP) Largest proxy dataset of codec-resynthesized speech against CodecFake; 4.3K+ downloads ·  HF
  • : DFADD (IEEE SLT) Mentored students on the first deepfake dataset from diffusion/flow-matching TTS; 9.1K+ downloads ·  HF
  • : SingGraph (INTERSPEECH) Achieved SOTA results on SingFake detection, fusing MERT and wav2vec 2.0; 24 GitHub stars ·  GitHub
  • : MTDVocaLiST (ICASSP) Compact audio-visual lip-sync model distilled from VocaLiST; 29 GitHub stars ·  GitHub

Collaborative & Community Work

  • : DeSTA2.5-Audio (IEEE TASLP, Core Contributor) 142 GitHub stars, HF model (28.7K+) and data (11.0K+ downloads) ·  GitHub
    • Curated training data for audio deepfake understanding as part of an audio LLM project in collaboration with NVIDIA
  • : Dynamic-SUPERB Phase-2 (ICLR, Task Editor) 201 GitHub stars, 325 HF datasets (289K+ downloads) ·  GitHub
    • Reviewed and standardized community-submitted tasks for the 180-task audio understanding benchmark
  • : Codec-SUPERB (ACL Findings & IEEE SLT, Core Contributor) 310 GitHub stars and a public leaderboard ·  Website
    • Built task- and signal-level metrics to benchmark several codecs on the neural audio codec leaderboard

Selected Honors and Service

See all on the Misc page.

Research and Academic Honors

Service and Outreach