The main purpose of this project is to minimize the time processing. The idea is very simple; first I apply a MFFC parameterisation of the speech signal. The result vector is mapped using Vector Quantisation method. At this code book I add the pitch parameter. All this process represents the training part of the program. In the recognition part the pitch parameter is very important. First we extract the pitch of the sample that we want to recognize and compare with the values that we obtained in training part. All the values in the training part that are equal with the pitch value of the sample used in the recognition part plus or minus e we validated. Then we applied the VQ method to recognise the speaker only for the remained samples who match. Using pitch as a first decision parameter I reduce the time processing with all most 40% (this is a relative percent and it depends by the number men or women who are in our base), because the time that we need to find the pitch value for a voiced signal is smaller then the MFFC parameterization of the same sample.
展开▼