Feature extraction, classification, and evaluation processes are considered the main concerns in speech recognition system and still the most active area of research nowadays. In this paper the performance of new hybrid feature extraction algorithm is examined using Linear Prediction Cepstrum Coefficient (LPCC), perceptual linear production (PLP), Mel Frequency Cepstrum Coefficient (MFCC), and RASTA-PLP. The extracted data vectors are classified and evaluated using Hidden Markov Model (HMM). The performance of the proposed hybrid algorithm is assessed using data set of human voice, which consists from eleven words (zero to nine plus oh) and recorded from 4558 adult speakers, each person said the same word 2 times. The collected data are sampled by 8 KHz and saved in 4558 WAV files divided into training and testing data. According to the final results, the proposed system provided an excellent recognition rate with 99.43% using the combination between MFCC and RASTA.