Chilin Shih

East Asian Languages and Cultures
The Beckman Institute
University of Illinois, Urbana-Champaign
Publications Courses Corpus Methods Prosody Presentations

I joined UIUC in 2003. I am affiliated with the Department of East Asian Languages and Cultures, the Department of Linguistics, and the Beckman Institute. Previously I was a speech scientist at Bell Labs, and before that, I taught phonology, linguistics and Chinese at Cornell University and Rutgers University. Currently, my work continues along the same theme: Chinese, linguistics, phonology, phonetics, and speech technologies. I worked on Text-to-Speech systems in many languages: Mandarin, Romanian, Russian, Hindi, Greek and Navajo. I also worked on some components of Spanish, Italian, and Japanese. My expertise is in acoustic inventory design, duration modeling, and intonation modeling. Over the years, we have collected speech databases on these languages that are excellent materials for the study of sound systems, coarticulation effects, and prosody.

My research includes Chinese linguistics, phonetics, phonology, as well as prosody modeling and generataion. The focus of my recent research interest is in prosody modeling. My works invovles the generation of Chinese tone and intonation, the cross-cultural communication and mis-communication of prosodic cues, and the role of prosody in second language acquistion.

I developed Stem-ML (Soft TEMplate Markup Language) with my colleague Greg Kochanski. This is a multi-lingual prosody modeling system that is suitable for both linguistic analysis and data-driven machine learning.

I gave a conference tutorial at ICSLP 2002 on Prosody and Prosodic Modeling . The tutorial materials are included in the link Prosody . It explains my philosophy of prosody modeling and the Stem-ML system. You can view Stem-ML as a phonetic implementation system. It is a generation system: given phonological category and a continuous parameter indicating the prosodic strength of the unit, it generates f0 contours deterministically. The information of phonetic variation is coded in the strength parameter. In the reverse direction, given phonological category and data, it finds the best strength values that account for the data.

I include my phonetics course materials in the link Phonetics . There are instructions to setup a basic, but functioning phonetic lab on your own laptop, using only free softwares. You can also find speech recordings of a Korean database and instructions to guide beginning students to complete a mini phonetic experiment.

In the Chinese pages, you can find materials of the Third Year Chinese Variety Show. This is a second language fluency training course design. I am very proud of my students' speeches and their talk show topics.