LPC-synthesis mixture: a low computational cost speech enhancement algorithm

The purpose of this paper is to develop a speech enhancement algorithm to reduce unwanted noise in a speech signal. The output of this enhancement system is a linear combination of the actual speech input to the system and a synthesized speech signal. The synthesized signal is generated using the LP coefficients and an estimate of the excitation from the current speech. The mechanical sound of the LP signal is further reduced by using an excitation signal based on the actual human excitation, not an artificially generated signal. This algorithm attempts to improve speech quality by emphasizing the speech signal rather than removing the background noise. The goal of the algorithm is to produce enhanced speech at telecommunications quality with little computational overhead.