The main purpose of this work was to find a suitable front-end analysis for Thai speech recognition by comparing the performance of the LPCC, MFCC and DCTC front-ends using several Thai continuous digit recognition tasks. HTK tools were used to build a word-based HMM recognizer which could handle several styles of digit string realization. Experimental results show that MFCC and DCTC perform equally well, and outperform LPCC. Dynamic parameters were also added to the feature sets to improve modeling accuracy. This work can be regarded as a baseline for further study of Thai speech recognition.