The objective is to build a “text dependent speaker identification/verification system.”
1. Record 5 utterances of the same “key phrase” (at least 0.5 second long, andcan be in any language). Collect the recorded data from all the students registered for this course.
2. Extract features using standard cepstra (EE), and LPCC (CSE/PRMLstudents) from ALL the utterances. Keep 3 sequences of feature vectors for every speaker as train templates. Use the remaining two for test.
3. Use time synchronous DTW to determine the identity of the speaker.CSE/PRML students perform score normalization (use Z and T norms), and obtain the EER. Others perform closed set identification, and produce a confusion matrix.
