Mei-Yuh Hwang (黃美玉)

http://ssli.ee.washington.edu/~mhwang/

Senior Research Scientist The Signal, Speech and Language Interpretation Lab Electrical Engineering, Room 215C University of Washington PO Box 352500 Seattle, WA, 98195 Tel: 206-221-5476 myhwang AT u DOT washington DOT edu

Mei-Yuh joined the SSLI Lab in March 2004, mainly responsible for developing Mandarin speech recognition systems using the SRI Decipher engine. Her main interests lie in pattern recognition (especially speech and handwriting recognition), statistic modeling, heuristic search, discrete math and algorithms.

Before coming to UW, she enjoyed 10 years at Microsoft Corporation , where she contributed to both research and product development which included the Whisper dictation system, Microsoft Speech API, Office XP English/Mandarin/Japanese dictation and Speech Server. She is familiar with SPHINX, CMU-LM Toolkit, SRILM Toolkit, HTK, and Microsoft and SRI Decipher speech engines. She has published many conference and journal papers and serves as a reviewer for IEEE Audio, Speech and Language Processing, Computer Speech and Language, and Speech Communication.

Research Projects at SSLI

• DARPA GALE Mandarin speech to English text translation, 9/2005-2/2008.
  • GALE PI meetings
  • Chinese audio format
  • Chinese text format
  • Unicode numeric references in XML
  • LDC GALE data progress
  • GALE MT Eval schedule
  • GALE SRI Team
• DARPA EARS Mandarin conversational telephone speech recognition, 3/2004 -- 8/2005.
  • RT04 Workshop
• Useful tools:
  • Basic reading on n-gram language modeling: Goodman technical report
  • N-gram training with SRILM Toolkit.
  • Chinese text processing:
    • To convert between different encodings, use iconv command:
      iconv -c -f FromEncoding -t ToEncoding -o OutputFile SourceFile
      The encoding names include (case sensitive)

      Encoding name	Explanation
      EUC-CN	GB2312
      EUC-TW	BIG5
      UTF-8	UTF-8

      For a list of all encodings, run iconv -l. The -c option omits invalid characters from output. See iconv help for more explanation.
    • Chinese word segmenter
      • Use N-gram (N=1,2,3) based maximum likelihood segmentation, or longest-first segmentation. (wseg.tar.gz)
      • Achilles word segmenter.
    • How to score character error rates? score.tar.gz
    • NYU Chinese Name Tagger
    • sox manual
    • gdb quick reference

  ---

  Publications

  Coming soon

  ---

  Education

  • Ph.D., Computer Science, Carnegie Mellon University, December 1993.
    Co-developing SPHINX-II, along with the CMU speech group.
    Thesis: Sub-phonetic Acoustic Modeling for Speaker-Independent Continuous Speech Recognition
    
  • B.S., Computer Science and Information Engineering, National Taiwan University, 1986.
    Phi Tau Phi Scholastic Honor Society, 1986.
    NTU Varsity Volleyball, 1982-1985.

  ---

  Curriculum Vitae

  Upon request.

  ---

  Personal homepage

  ---

  Back to the SSLI Lab home page.

  ---

  Copyright ©1995-2002 University of Washington, College of Engineering - SSLI Lab
  E-mail any questions/comments/corrections to: mhwang@ee.washington.edu