Towards a minimally redundant linear predictive vocoder