Abstarct: Speaker identification is one of most important branches of speech recognition which has many applications in speech-baxsed secure systems. In last decades, many efforts have been done in order to improve performance of such systems. Much of these efforts have focus on improving recognition rate and have not paid much attention on other important parameter such as hardware implementation and being online. This thesis concerns to online implementation of a speaker identification algorithm on TMS320C5509A DSP processor platform. After intensive investigation, we choose spectral short time methos, especially MFCC and LPCC for feature extraction and Gaussian Mixture Model (GMM) for speaker modeling. Simulation results in MATLAB show that MFCC method has higher recognition rate than that of LPCC. Although MFCC has 99% recognition rate in noiseless conditions, but in presence of noise it degrades dramatically. In addition, learning GMM models in noisy conditions decreases the recognition rate noticeably. Because of sensitivity of MFCC against noise, we used TIMIT noiseless databaxse. In addition, we had not capability to generate a noisy standard. In hardware experiments, we directly loaded the program on DSP processor using JTAG cable. Results show that the algorithm can be performed online. Maximum recognition rate was obtained as 78.6%.