Mean Teacher for Weakly Supervised Polyphonic Sound Event Detection: An Empirical Study