Postdoctoral Researcher, MIT CSAIL

Kai-Wei Chang 張凱爲

I focus on speech research in full-duplex spoken dialogue systems and low-resource speech technologies.

kwchang at mit dot edu CV Google Scholar GitHub LinkedIn
Kai-Wei Chang

About

I am a postdoctoral researcher in the Spoken Language Systems Group at MIT CSAIL, working with Dr. James Glass. I received my Ph.D. in Communication Engineering from National Taiwan University, advised by Prof. Hung-yi Lee.

My research has two main threads: (1) full-duplex spoken dialogue systems, where I study temporal dynamics in full-duplex conversations, and time awareness in real-time voice agents; and (2) low-resource speech technologies, using Taiwanese as a concrete example for building useful speech systems when data, writing systems, and deployment conditions are limited.

I built public-facing Taiwanese speech and language tools. I created TaigiTube, a Taiwanese learning platform based on contextualized YouTube search, and lead TaigiSpeech, a real-world Taiwanese speech dataset project for eldercare voice assistants. These projects have been covered by major Taiwanese media, including TVBS, PTS, FTV, and BCC.

My dissertation, Towards a Universal Speech Model: Prompting Speech Language Models for Diverse Speech Processing Tasks, received the ACLCLP Best Dissertation Award in 2025.

Current Postdoctoral Researcher, MIT CSAIL, July 2025 - June 2027
Education Ph.D., National Taiwan University, 2025; B.S., National Taiwan University, 2020
Contact kwchang at mit dot edu

News

  1. Our paper Game-Time: Evaluating Temporal Dynamics in Spoken Language Models is accepted by ICASSP 2026.
  2. Our paper TaigiSpeech: A Low-Resource Real-World Speech Intent Dataset and Preliminary Results with Scalable Data Mining In-the-Wild is accepted as an INTERSPEECH 2026 long paper.
  3. I am organizing and speaking in the INTERSPEECH 2026 tutorial On the Landscape of Spoken Language Models.
  4. Our paper BILLY: Steering Large Language Models via Merging Persona Vectors for Creative Generation is accepted by EACL 2026.
  5. I gave the invited talk From TaigiTube to TaigiSpeech: Developing Taiwanese AI for Language Learning and Home Care Assistance at Stanford.
  6. I gave the invited talk Towards "Better" Dynamics in Full-Duplex Spoken Language Models at The University of Texas at Austin.
  7. I started as a postdoctoral researcher at MIT CSAIL.
  8. I launched TaigiTube, a Taiwanese-learning platform that reached 100,000 users within two weeks.
  9. Our survey On the Landscape of Spoken Language Models: A Comprehensive Survey is published in TMLR 2025.

Publications

Full list on Google Scholar and in my CV.

Selected First-Author Publications

  1. TiCo: Time-Controllable Spoken Dialogue Models Kai-Wei Chang*, Wei-Chih Chen*, En-Pei Hu, Hung-yi Lee, James Glass arXiv preprint
  2. Overcoming State Inertia in Full-Duplex Spoken Language Models via Activation Steering Cheng-Kuang Chang*, Kai-Wei Chang*, Alexander H. Liu, James Glass arXiv preprint
  3. Game-Time: Evaluating Temporal Dynamics in Spoken Language Models Kai-Wei Chang*, En-Pei Hu*, Chun-Yi Kuan, Wenze Ren, Wei-Chih Chen, Guan-Ting Lin, Yu Tsao, Shao-Hua Sun, Hung-yi Lee, James Glass ICASSP 2026
  4. TaigiSpeech: A Low-Resource Real-World Speech Intent Dataset and Preliminary Results with Scalable Data Mining In-the-Wild Kai-Wei Chang, Yi-Cheng Lin, Huang-Cheng Chou, Wenze Ren, Yu-Han Huang, Yun-Shao Tsai, Chien-Cheng Chen, Yu Tsao, Yuan-Fu Liao, Shrikanth Narayanan, James Glass, Hung-yi Lee INTERSPEECH 2026, long paper
  5. On the Landscape of Spoken Language Models: A Comprehensive Survey Siddhant Arora*, Kai-Wei Chang*, Chung-Ming Chien*, Yifan Peng*, Haibin Wu*, Yossi Adi, Emmanuel Dupoux, Hung-Yi Lee, Karen Livescu, Shinji Watanabe TMLR 2025
  6. SpeechPrompt: Prompting Speech Language Models for Speech Processing Tasks Kai-Wei Chang, Haibin Wu, Yu-Kai Wang, Yuan-Kuei Wu, Hua Shen, Wei-Cheng Tseng, Iu-thing Kang, Shang-Wen Li, Hung-yi Lee IEEE/ACM TASLP 2024
  7. Exploring In-Context Learning of Textless Speech Language Model for Speech Classification Tasks Kai-Wei Chang*, Ming-Hao Hsu*, Shang-Wen Li, Hung-yi Lee INTERSPEECH 2024
  8. Prompting and Adapter Tuning for Self-Supervised Encoder-Decoder Speech Model Kai-Wei Chang, Ming-Hsin Chen, Yun-Ping Lin, Jing Neng Hsu, Paul Kuo-Ming Huang, Chien-Yu Huang, Shang-Wen Li, Hung-yi Lee ASRU 2023
  9. An Exploration of Prompt Tuning on Generative Spoken Language Model for Speech Processing Tasks Kai-Wei Chang, Wei-Cheng Tseng, Shang-Wen Li, Hung-yi Lee INTERSPEECH 2022
  10. Toward Degradation-Robust Voice Conversion Chien-Yu Huang*, Kai-Wei Chang*, Hung-Yi Lee ICASSP 2022

Selected Last-Author Publications

  1. BILLY: Steering Large Language Models via Merging Persona Vectors for Creative Generation Tsung-Min Pai, Jui-I Wang, Li-Chun Lu, Shao-Hua Sun, Hung-Yi Lee, Kai-Wei Chang EACL 2026, corresponding author

* Equal contribution.

Press

From TaigiTube to TaigiSpeech

I started TaigiTube as a Taiwanese-learning platform that uses contextualized YouTube search to make Taigi learning more lively and connected to everyday media. The project reached 100,000 users within two weeks and led to interviews from TVBS, PTS, FTV, and BCC. The effort grew into TaigiSpeech, a broader research project to build real-world Taiwanese speech resources for AI, including an eldercare voice-assistant dataset with more than 3,000 utterances from over 20 older adult speakers (paper accepted to INTERSPEECH as a long paper).

BCC Radio

In-depth 40 min personal interview

BCC YouTube

Talks

Mentoring

  • Mentored over 20 undergraduate and graduate students at MIT and NTU, each working closely with me for at least one semester on research that led to a publication or industry collaboration.
  • Mentee outcomes include graduate study in CMU ECE (USA), CMU CS (USA), UT Austin ECE (USA), CUHK Shenzhen Ph.D. (China), ETH Zurich (Switzerland), and NTU Ph.D./M.S. (Taiwan) programs, as well as roles at Google US, Google Munich (Germany), Amazon US, Uber US, TSMC Taiwan, CyberLink Taiwan, and ASUS Taiwan.

Teaching

  • Teaching Assistant, Deep Learning for Human Language Processing, Fall 2022 and Spring 2022
  • Head Teaching Assistant, Machine Learning, Spring 2021; coordinated more than 40 TAs for a course with more than 1,300 students
  • Head Teaching Assistant, Linear Algebra, Fall 2020; coordinated six TAs for approximately 130 students

Professional Services

  • Program Chair, ROCLING 2025
  • Publication Chair, ISCSLP 2026 and ROCLING 2025
  • Session Chair, ICASSP 2026 and ICASSP 2025
  • Tutorial organizer and speaker, INTERSPEECH 2026
  • Tutorial speaker, ICASSP 2024 and ICASSP 2023
  • Challenge co-organizer, Codec SUPERB Challenge at SLT 2024
  • Area Chair / Meta-Reviewer, SLT 2026, ASRU 2025, ICASSP 2026
  • Reviewer for TASLP, TPAMI, NeurIPS, INTERSPEECH, ICLR, ICASSP, IJCNN, ARR, MLSP, SLT, IALP, and ACML