Controlling 'shout' expression in a Japanese POP singing performance: analysis and suppression study

摘要

Degree of "shout" singing performance is effectively controlled by combining global spectral shape equalization, peak cancellation in frequency modulation spectrum of FO trajectory, and synchronized shape-modulation of voice spectral envelope. This "shout-reduction" processing is based on a symmetry- based FO extractor with fine temporal resolution, a temporally stable representation of instantaneous frequency of periodic signals, and the TANDEM-STRAIGHT, a speech analysis, modification and resynthesis framework. The proposed procedure successfully converted an expressive Japanese POP song performance with "shout" into a plain performance without damaging original naturalness. Possibility of adding artificial "shout" to plain performance is also discussed.