Applications of Robust Estimators in Speech Signal Processing

It has been shown in previous chapters that using the robust methods of linear prediction, it is possible to determine the parameters of the AR model of the system for speech signal production with a greater accuracy. The performed analysis has shown that in this way it is possible to more accurately model the vocal tract. Robustification of non-recursive LP methods ensures lower sensitivity of estimations to the fundamental speech frequency, as well as their lower sensitivity to the length and position of the analysis interval.