Humming Jukebox A human nature - Query by Humming
People Research Showcase Resource Home
Hsuan-Huei (Maverick) Shih
Ph.D. Candidate
Department of Electrical Engineering

Email:hshih@sipi.usc.edu
Phone: (213)740-0468
Advisor: Prof. C.-C. Jay Kuo
Co-advisor: Prof. Shrikanth S. Narayanan

 
Maverick Shih received the B.S. degree of Apply Math from Fu-Jen University, Taiwan in 1995, and M.S. degree of Electrical Engineering from University of Southern California, U.S. in 2000. He is currently a Ph.D. Candidate in the Department of Electrical Engineering, University of Southern California. His research interests are in the areas of audio content analysis, multimedia database management, preprocessing of Query by Humming system, and humming recognition.

Research Interest:

My research is focus on preprocessing part of a Query by Humming system. Query by Humming is a system which uses human humming to do the query in music database.
Audio content analysis and multimedia database management are two of important preprocessing of query by humming system.

Audio content analysis:

Transferring acoustic waveform to symbolic form is a difficult task especially for human humming. Before doing any query process, humming needs to be converted into symbolic format for matching process. Due to the nature of human humming, inaccurate pitch and big variation between users, humming recognition become an interesting and challenging research topic. I tackle this problem using statistic approach.

Multimedia database management:

Doing linear search within a huge music database is not a very good approach, although it will return a desired result, however it usually required a lot of time. Music database management becomes very important to have a well-organized music database. There are features in music which can be used to organizing music database, for example: title, artists, music type, or repeating melodies. Repeating melodies are the keys of a music piece, and they are easily memorized by people. Extracting repeating melodies from a piece of music can help the management to have better organized music database.

Development tools:

  • HTKEdit: A GUI (Graphic User Interface) for HTK (Hidden Markov Model Toolkit), written in Visual C++
  • HummingDecoder: A GUI Humming Decoder using the latest note models, written in JAVA
  • Repeating Pattern Finder: A GUI program which finds repeating patterns in MIDI files

  • Home | Sitemap | Contact us
    Copyright©2002. All rights reserved.