The work aims at recognizing words from a continuous speech. To achieve this, cepstrum analysis of the speech signal is carried out. The speech signal is processed and the features are extracted using cepstrum analysis. The extracted features are given as inputs for the hidden Markov model (HMM) followed by training radial basis function (RBF). During the testing process, the words are separated and compared in the database. If a word matches then subsequent action is carried out. If the word is not present, then it is added to the database.