FFT analysis gives you the frequency components but you miss the loaclized time properties. You can make use of wavelet decomposition and look at the features at different frequency sub-bands. Cross correlation is also a good tool, but you have to be good at interpreting the results.. --- In , "Rick Lee" <rlee@a...> wrote: > Dear All, > > I am doing a project for speech recognition, which method will be > good to compare two speechs and tell how much they are matched? > > I am trying using LPC and FFT, but the result is not so good, maybe I > should do some quantiztion first, any idea? > > Thanks a lot! > > Rick |