Speech analysis and synthesis using a glottal excited AR model with DTW-based glottal determination

In this paper we present, a new method for determining the excitation waveform of a glottal excited speech synthesizer. In the glottal excited speech model, voiced speech is decomposed into a parametric glottal shaped excitation signal and an AR vocal tract filter. The new glottal determination method can be used iteratively with vocal tract filter updates in model analysis. In each iteration the new glottal pulse positioning is determined at sub-sample resolution using by a Multi-Dimensional Dynamic Time Warping (MD-DTW) algorithm, using an analysis-by-synthesis approach to minimize the squared-error between the original and the synthetic speech.