Abstarct: Keyword spotting is an important branch in speech signal processing which has many application in security and human-intraction-baxsed systems. Literature review shows that the main effort has been done on improving the identification accuracy of spoken keywords not the hardware implementation of an approprate approach from aspects such as real-time processing, low power consumption, etc. The aim of this thesis is to hardware implementation of an approprate method baxsed on the TMS320C55xx digital signal processors. After reviewing the litrature , we identified the short-time spectral methods as approprate family and among which, we considered the MFCC and LPCC approaches to be implemented on hardware. Simulation resulte show the MFCC approach result in better performance compared against the LPCC one. We simulated the automatic keyword spotting system in two modes of discrete and continuous speech. The average identification accuracy, when we used individually spoken numeral keywords, was computed as 93.33% and 81.58% in the discrete and continuous modes, respectively. In addition, the average identification accuracy, when we used keywords from spoken speech, was computed as 70.36% and 70.66% in the discrete and continuous modes, recpectively.