automatic generation of karaoke-game stage presentation at nsc
DESCRIPTION
Automatic Generation of Karaoke-Game Stage Presentation at NSCTRANSCRIPT
Automatic Generation of Karaoke-Game Stages from Audio Files
Natthawut KulnirundornProfessor Boonserm Kijsirikul
Department of Computer EngineeringChulalongkorn University
1/9/2009
1
Project Background• Karaoke Game
▫ User try to sing at correct pitch for each note
▫ Can analyze how well you sing
• People enjoy playing karaoke game
▫ Existing commercial products: SingStar, Rockband
1/9/2009
2
How to play karaoke game?
54100
1/9/2009
3
Problems with current karaoke games• Limited number of stages
▫ Usually pre-packed with game
▫ Cannot add new song
• Available on limited platforms▫ Singstar on PS3▫ Rockband on XBOX360
Need to buy new game to play more songs!
1/9/2009
4
Proposed Solution
1/9/2009
5
Project Objectives
1/9/2009
6
Vocal Pitch Detection in Polyphonic Audio• Still an open research
problem
• Cannot simply apply the same algorithms as pitch detection in single source
▫ Autocorrelation
▫ Harmonic Product Spectrum (HPS)
• Current approaches based on many different techniques
▫ Machine Learning
▫ Independent Component Analysis (ICA)
Detected frequency from HPS
Freq. from Human voice highlighted in red
1/9/2009
7
timefreq
Demonstration – Stage Generation
1/9/2009
8
Demonstration – Karaoke Game
1/9/2009
9
Method Proposed
1/9/2009
10
Spectrum Subtraction
1/9/2009
11
normal
voice-removed
result
time
freq
*Human singing frequency
Conclusion
•Enable users to sing a song in native languages.
•The method proposed is easy to implement and can be applied to classes of application
•The prototypes can be developed further to be an entertainment application
1/9/2009
12
Questions ?
1/9/2009
13
References
1/9/2009
14
Appendix1/9/2009
15
Autocorrelation
1/9/2009
16
Harmonic Product Spectrum
1/9/2009
17