This study aims to build a speaker recognition system using the Discrete Wavelet Transform (DWT) algorithm and Hidden Markov Models (HMM). Speech signals from each speaker were recorded using Indonesian words "kiri" and recorded 10 times. Five (5) data from the first record were recorded under normal conditions and the next 5 data were sourced from the nasal sounds produced by a pressed nose. The total data from 6 different speakers becomes 60 data The results of the application of the Discrete Wavelet Transform (DWT) algorithm and the Hidden Markov Models (HMM) algorithm and the number of states tested 4 to 7 states, in this study have not provided optimal results. The identification of error rates is quite high, which is equal to 20% for the number of states 4 and 5, and reaches 30% for the number of states 6 and 7. This shows that the feature vector values generated from the DWT algorithm and then modeled and tested using the HMM algorithm has not optimal results yet. Further evaluation is needed to examine the opportunities of other algorithms that can be applied in DWT and to achieve high recognition accuracy