Abstract: In this paper, an improved strategy for automated text dependent speaker recognition system has been
proposed in noisy environment. The preprocessing of speaker signal started with eliminate the background noise.
The next step is signal filtering and features extraction using cepstrum coefficients method, this extracted features
can be used to by the enhanced LBG for vector quantization algorithm for speaker recognition, such that the
specified speaker can be determined by matching the speaker to be tested with in stored codebook in database. And
finally select correct speaker that have the lesser Euclidean distance. The speech feature extraction was based on a
dataset of 175 different samples collected from 25 different speakers The results of the proposed system approved
with good recognition ratio of speaker identification with maximum accuracy about 96.2% for database with close
set of selected words contains the most used phonemes. Also the results of experiments show that recognition
accuracy increased with frames overlapping