Speech analysis and synthesis based on ARMA lattice model.

The research documented in this thesis presents a pitch synchronous speech analysis and synthesis system using an ARMA (auto regressive moving average) lattice filter. The LP (linear predictive) model is, by far, the most widely used speech production model in speech processing. A well-known drawback with the LP model is that it ignores the effect of the nasal cavity. The mismatch between the model and the vocal tract becomes severe for nasal sound production and leads to poor synthesized speech. The nasal tract effect can be included in the speech production model by representing the vocal tract with an ARMA filter. Thus, better synthesis speech is expected, at least for nasal phonemes. In this thesis, the ARMA lattice filter is used to model the vocal tract. Based on analysis and experiments, it is concluded that the estimation of excitation source for voiced speech is essential in obtaining the ARMA lattice filter coefficients and important in generating high quality synthesis speech. By comparison, LF model is chosen for this purpose. In the proposed analysis system, the analysis parameters include: ARMA filter coefficients, LF model parameters, voice type, pitch and gain, which are analyzed pitch synchronously. Correspondingly, the pitch synchronous speech synthesis system is developed. At the end, the proposed system is simulated for both nasal and non-nasal phonemes and the results are compared with those from the LP model.Dept. of Electrical and Computer Engineering. Paper copy at Leddy Library: Theses & Major Papers - Basement, West Bldg. / Call Number: Thesis2003 .W36. Source: Masters Abstracts International, Volume: 42-02, page: 0652. Adviser: H. K. Kwan. Thesis (M.A.Sc.)--University of Windsor (Canada), 2003