Hi, I am new here and this is my first use of WEKA, and I really need help with my task

I've got series of recorded audio samples with spoken sentences. I am using "Praat" for extraction parameters of human voice (like pitch, formant frequencies ect.), this process is automated by macro i wrote. I'm exporting data as .csv files containing table of parameters in time points. Now i need somehow to use WEKA and build learning model for comparison these sentences. There are 7 samples in each category, and 3 of them are correct and needs to be used as training set.

My data file looks like this (at this moment):
First of all, i don't know how to save data correctly, i see two options:
1) export data set for each sample in separately .csv file, which means there will be 7 .csv files, 3 to built training model and tested on all 7 files
2) export training data in one file (data from 3 samples exported as one csv) and test data in one (7 in one), or test data separately?
I don't know how to do this correctly because in experimenter i am able to load multiple data files, in explorer just one + test set
I don't know how to use WEKA, i have tried do this witch explorer, experimenter, i found info about serializedClassifier but... i am totally lost, there are too many options, filters, classifiers, error massages, secret bonuses and magic things.

My job is to build learning machine working on data extracted from samples. This machine have to compare whole set of samples (7) to model trained on 3 'correct' samples. The point is to get results which shows something like probability of correctness for each sample, degree of accuracy, etc.

I'm not asking for whole solution, but if someone can show me the way, i would be very thankful. I have only few days for this task, and I really need help.

Some ideas?