Analysis and encoding of speech for a parallel formant synthesizer

The authors develop a robust algorithm for formant amplitude estimation using analysis-by-synthesis techniques. The results of this study indicate the possibility of a viable alternative to the complex and time-consuming JSRU analysis procedure for the parallel formant synthesizer. The proposed amplitude estimation algorithm incorporates a more accurate software model of the synthesizer in an analysis-by-synthesis loop and reduces complexity by adopting a fixed frame rather than a pitch synchronous approach. Tests with synthetic speech have shown that essentially the same set of amplitudes can be recovered by this method, and good quality speech has been obtained by analysis of natural speech. Further refinement of the algorithm is required to treat special cases such as merged formants and to reduce the effects of occasional errors due to pitch pulse misalignment