| Previous | [ 1] | [ 2] | [ 3] | [ 4] | [ 5] | [ 6] | [ 7] | [ 8] | [ 9] | [ 10] | [ 11] | [ 12] | [ 13] | [ 14] | [ 15] | [ 16] | [ 17] | [ 18] | [ 19] | [ 20] | [ 21] | [ 22] | [ 23] | [ 24] | [ 25] |
¡@
HUNG-YAN GU AND HUANG-LIANG LIAO
Department of Computer Science and Information Engineering
National Taiwan University of Science and Technology
Taipei, 106 Taiwan
In this paper, HNM (harmonic plus noise model) is enhanced and used to design a
scheme for synthesizing a Mandarin Chinese singing voice. Enhancements made include
a Lagrange-interpolation based estimation of spectral envelope, piecewise linear mapping
of time axes, fixed-pace placement of control points, and other modifications for analyzing
HNM parameters and efficient execution. In terms of the enhancements and the signalsynthesis
equations rewritten here, a Mandarin singing-voice synthesis system is built. In
the system, each Mandarin syllable is recorded just once for analyzing HNM parameters.
Then, the HNM parameters of a source syllable are used to synthesize singing syllables of
diverse pitches and durations. This system can parse a song score file and synthesize its
lyric syllables¡¦ signals in real-time. Also, the skill of portamento (pitch gliding) singing is
implemented. According to the perception tests, our system can indeed synthesize signals
of singing voice that are consistent in timbre, of no reverberation, and much clearer than
a PSOLA (pitch synchronous overlap add) based scheme.
Received January 5, 2009; revised April 21 & June 22, 2009; accepted August 13, 2009.
Communicated by Chin-Teng lin.
* This research was partially supported by National Science Council of Taiwan under Grant No. NSC-95-2218-
E-011-009.