2-PHASE UNWRAPPING CRITERIA FOR QUADRATIC AND CUBIC INTERPOLATION IN VOICED SPEECH SYNTHESIS

In this paper, two-phase unwrapping criteria appropriate to different conditions for voiced speech synthesis are suggested. When the synthes is frame lags the analysis frame by half a frame, the first one, named Criterion A, should be used. It leads to the same unwrapping results when it is used for both quadratic and cubic interpolation. In additio n, for quadratic interpolation, Criterion A is equivalent to the crite rion to minimize the absolute frequency deviation, suggested by Griffi n and Lim (1988); for cubic interpolation, it is equivalent to the 'ma ximally smooth' criterion, proposed by McAulay and Quatieri (1985). Ho wever, when the synthesis frame coincides with the analysis frame, the second criterion, named Criterion B, should be used. It also leads to the same unwrapping results for both interpolation methods. In the se cond case, simulations using the multi-band excitation (MBE) model to synthesize speech have shown that Criterion B outperforms Criterion A.