Spectral analysis of speech signals in noisy environments is an aspect of signal processing that deserves more attention. This study uses some of the recently proposed robust Linear Prediction (LP) analysis methods for estimating the power spectrum envelope of speech signals. These methods provide robustness in spectral estimation without sacrificing their accuracy. These methods, which are the Moving Average (MA), Moving Maximum (MM) and the Average Threshold method (AT), are compared to the more commonly used methods of LP analysis, such as the widely used autocorrelation method and the Spectral Envelope Estimation Vocoder (SEEVOC) method. The LP power spectrum envelope estimates from these methods are found to be more efficient for speech coding applications as well as more robust to noise.