1 Filter the vad

-   by running vad_processing/vad_processing.ipynb
-   fitered vad files should be put in the path preprocess/audio/filter_vad/

2 Using the filtered vad files to generate the samples, ground truth label and write them into csv files

-   by running preprocess/audio/generate_samples.py 
    
    #main(x, y, z, vad_dict) 
        -   x : experiment num (0 indicates generating training samples)
        -   y : window size
        -   z : ratio of positive samples to negative samples

-   make sure generate training samples first.

3 Making pkl files for corresponding experiments' samples

-   by running data_loading/make_examples.py

    -   generate training samples  
        -   make_all_examples(0, windowSize)

    -   generate samples for experiment 1
        -   make_all_examples(1, windowSize, numberOfExperiment)

    -   generate samples for experiment 2
        -   make_all_examples(2, windowSize, numberOfExperiment)

    -   generate samples for experiment 3
        -   make_all_examples(3, windowSize, numberOfExperiment, 'all_unsuccessful')
    
    -   generate samples for experiment 4
        -   make_all_examples(4, windowSize, numberOfExperiment, 'start')

    -   generate samples for experiment 5
        -   make_all_examples(5, windowSize, numberOfExperiment, 'continue')


### make sure the windowSize is consistant with the samples generated by generate_samples.py

4 Execute the training

-   running baseline/testTrain.py

    -   train the model (make sure you also have one successful test sample pkl with same windowSize )
        -   main(True, 0, windowSize, numberOfExperiment) 
        -   model is saved in 108th line
            -   torch.save(model.state_dict(), "savedName.pt")
    
    ## Choose the correct model for following experiments' test(correct windowSize)
    ## modeled is chosen in 122th line
        -   model.load_state_dict(torch.load("chosenModel.pt"))

    For experiment 1 - 5
        -   main(False, x, y, z)
            -   x : experiment number
            -   y : windowSize
            -   z : how mnay experiment run