(puzzled)

One needs to understand that the algorithm developed in this project are based on single voiced phonemes. As a future work these algorithms need to be generalized to multi-phonemes => words => sentences => conversations.  Furthermore, the extension of spectral envelope algorithm is based on probabilistic measures. It might be prudent to reconsider the actual ranges for the probabilistic parameters which may deal with changing the decoding mechanism itself. Also more  robust intermediates such as HFCCs or MFCCs should be used instead of the LPC coefficients, which are more sensitive to noise. If these algorithms are to be integrated onto a DSP then complexity issues need to be reconsidered.