Iterative inverse filtering by lattice filters for time-varying analysis and synthesis of speech

In this contribution, an analysis procedure is proposed for time-varying analysis and synthesis of speech based on lattice filters. The estimation is performed by an iterative inverse filtering approach exploiting analytical suboptimal solutions. Starting from an initial configuration of coefficients, the procedure estimates a continuous piece-wise linear trajectory in terms of reflection coefficients. In this way, smooth trajectories can be estimated which have additionally a high time resolution. One advantage of this analysis technique is that the coefficient trajectories are estimated in the same way as they are used for the synthesis. Examples of synthesized speech signals show that the proposed algorithm suppresses artifacts which are caused by the use of time-invariant estimation procedures.