Joint linear prediction and epoch estimation of voiced speech using a basis where the prediction residual can be sparsely represented

Whereas most approaches to linear speech prediction fail to account for the quasi-periodic glottal flow, this paper incorporates the Liljencrants-Fant model for glottal flow derivative (GFD) directly into the linear prediction problem. A linear model for the prediction error is obtained by constructing a dictionary of time-shifted GFD pulses. Minimizing the difference between the linear prediction residual and a sparse combination of the pulses in the dictionary leads to joint estimation of the linear predictor as well as a sparse representation for the prediction error that reveals the instants of vocal tract excitation (epochs). A greedy algorithm is proposed to approximately solve this joint estimation problem. The method is applied to voiced segments extracted from the CMU Arctic dataset which also includes electro-glottograms. Results show that the approach and the proposed algorithm are effective in estimating the parameters of interest.