Circular LPC modeling and constant pitch transformation for scalable bit rate, scalable quality speech coding

In this paper, circular LPC analysis, a windowless signal modeling method for periodic signals, is re-visited as a high-resolution pitch-synchronous speech spectrum analysis tool for high quality speech coding, and the constant pitch transform is re-introduced as an alternative representation for the (circular) residual signal. This combination results in a parametric speech representation that is uniquely well suited for many speech coding applications. It is shown that circular LPC requires a very accurate pitch period estimation. Thus, both techniques are formulated using an oversampled version of the speech signal for analysis. Experiments show that artifact free speech can be synthesized using the proposed method.