Low complexity robust linear predictive coding of speech signals

Abstract

Numerous authors have developed digital techniques for enhancing acoustically noisy speech signals. These algorithms are often applied as preprocessors to narrowband digital speech communications systems. Among the drawbacks of the pre- processor approach is the relatively high overall computational complexity. A method, using smoothed spectral estimates, is described for embedding many of these techniques within the autocorrelation approach to linear predictive analysis. This method provides a significant reduction in overall complexity.