Zhen Wan

I am a PhD student in Informatics (Intelligence Science & Technology) at Kyoto University, advised by Prof. Sadao Kurohashi and Prof. Chenhui Chu. I am also a research intern at NVIDIA Research, working with Dr. Chao-Han Huck Yang, Dr. Rafael Valle, and Dr. Boris Ginsburg on voice agentic and omni-perception models.

My research is on voice agentic systems and multimodal LLMs — specifically, teaching omni-models when to trust themselves versus when to consult external perception tools, and how to evaluate the intelligence of speech LLMs. Recent work includes:

• Speech-Hands (ACL 2026, Oral) — a self-reflection voice agentic framework for ASR and audio reasoning.
• SpeechIQ (ACL 2025 Main) — an agentic intelligence quotient for voice-understanding LLMs across cognitive levels.
• NeKo (ACL 2025, Best Industrial Paper Honor Mention) — cross-modality post-recognition error correction with mixture-of-experts.

I am supported by a JSPS DC2 fellowship (acceptance rate 19.1%). Before Kyoto, I received a B.E. in Energy Engineering from Zhejiang University. Outside research, I have been a research assistant on LLM-jp and the SIP medical LLM project, and previously interned at Alibaba DAMO Academy on Chinese legal LLMs.

I regularly review for ACL, EMNLP, NAACL and ARR. Feel free to reach out — I am happy to chat about voice agents, multimodal LLMs, low-resource Japanese NLP, or anything in between.