Mandarin syllable signal synthesis using an HNM based scheme

Abstract

In this paper, HNM (harmonic plus noise model) is enhanced and used to design a scheme for synthesizing Mandarin syllable signals. Each syllable is recorded once only and used to synthesize syllable signals with diverse prosodic characteristics without suffering significant signal-quality degradation. For a control point on the synthetic syllable's time axis, two corresponding analysis frames' HNM parameters are interpolated to derive the HNM parameters for the control point. Furthermore, for pitch-contour tuning, another timbre-reserving interpolation is performed for the HNM parameters on a control point. Then, signal samples are synthesized with the HNM synthesis equations rewritten here. According to the result of the perception tests, the HNM based scheme proposed here can indeed be used to synthesize syllable signals with consistent timbre and high signal clarity.