According to Mergen logmel and entire audio preprocessing should be possible to do inside ET model, thus removing one dependency from our library.