Noise immune speech recognition system

This paper investigates the performance of an isolated word recognition (IWR) system in a noisy environment. Two approaches have been demonstrated to overcome the effect of the noise on the recognition accuracy. These approaches are, using noise immune features and reference model contamination. The performance is evaluated in a noisy environment at different signal-to-noise ratios (SNR), with different feature extraction techniques including linear predictive coding (LPC), cepstrum analysis, weighted cepstrum analysis, and perceptual linear predictive coding (PLP). The performance of these features is compared based on the recognition accuracy. The results have shown that the PLP features exhibits the best noise immunity and recognition accuracy among the studied features