Abstract : This paper presents robust speech recognition system in the presence of noise. Discrete Hidden Markov Model (DHMM) is used for mainly reducing the computation burden of voice recognition which in turn increases speed. Hilbert Huang Transform (HHT) is an empirical approach to decompose any complicated data set into a finite number of Intrinsic Mode Functions (IMF) to obtain the instantaneous frequency data. This Empirical Mode Decomposition (EMD) method of HHT operates in time domain on the local characteristic time scale of the data, making it adaptive and highly efficient to work with any nonlinear and nonstationary data’s unlike Fourier transforms. The Mel Frequency Spectrum Coefficients (MFCC) is derived from cepstral coefficients of IMFs. The features are then weighted and summed to get the original speech reconstructed signal. Genetic Algorithm (GA) was designed for each IMF to get better optimal solution. This results in significant reduction in time measurement, and thus it improves the speech recognition rate