Sixth International Conference on Spoken Language Processing
(ICSLP 2000)

Beijing, China
October 16-20, 2000

Model Based Voice Decomposition Method

Masahide Sugiyama

Graduate School of Computer Science and Engineering, The University of Aizu, Aizu-Wakamatsu, Fukushima, Japan

This paper proposes a voice decomposition method based on voice model. Using correlation distance as a criterion frequency domain decomposition is formulated. For a pair of correlation coe cient vectors minimizing powers of spectra is characterized using quadratic programming problem with constraint, where values of powers are non-negative. The solution of the problem without constraint is calculated using a linear equation where the coecients are determined by correlation coecient vectors. A correspondingmatrix for the linear equation has a nonnegative determinant, and the sucient and necessary condition for existence of reverse matrix is that given correlation coecient vectors are linear independent. The solution with constraint is also given by solving a linear equation.


Full Paper

Bibliographic reference.  Sugiyama, Masahide (2000): "Model based voice decomposition method", In ICSLP-2000, vol.4, 684-687.