3 ms·

Note that WaveNet was not trained "in the wild" (like, on celebs) but rather on a speech style dataset