This csv is contain the first 8 channels
columns = 128 (1 sec)
rows = 116(seconds) * 8(channels) = 928
Each row as 128 "time samples" from 1 channel that convert to frequency by fft
batch size = 2 rows = 2 channels (Important: not all channel just 128 samples from it)
num of batches = rows / 2 = 464
Each source will get 116 freq _samples = 116 rows (2 full channels) or 464/4 = 116 (take 2 channels from all the batches)