Training a binary classifier (in case of 2 speakers)

[ukemamaster]

@joonson
To train a binary classifier (having 2 speakers in the entire data), what should be the values for max_seg_per_spk , nClasses, nPerSpeaker, and batch_size ?

i have been trying different values, but it seems like the datasampler works only for max_seg_per_spk = any_value, nClasses = 2, nPerSpeaker = 2, and batch_size = 1 or 2