A new approach for extracting significant characteristic within speech sign
al for distinct speaker is presented. Based on the multiresolution property
of wavelet transform. quadrature mirror filters (QMFs) derived by Daubechi
es is used to decompose the input signal into varied frequencies channels.
Owning to the uncorrelation property of each resolution derived from QMFs.
Linear Predict Coding Cepstrum (LPCC) of lower frequency region and entropy
information of higher frequency region for each decomposition process are
calculated as tile speech feature vectors. In addition, a hard thresholding
techniques fur lower resolution in each decomposition process is also used
tc, remove the effect of noise interference. The experimental result shows
that by using this: mechanism not only effectively reduce tile effect of n
oise inference but improve tile recognition I ate. The proposed feature ext
raction algorithm is evaluated on MAT telephone speech database for Text-In
dependent speaker identification using vector quantization (VQ). Some popul
ar existing methods are also evaluated for comparison in this paper. Experi
mental results show that the performance of the proposed method is: more ef
fective and robust than that of the other existing methods. Fur 80 speakers
and 2 seconds utterance, the identification rate is 98.52%. In addition, t
he performance of our method is vcr satisfactory even at low SNR.