A Closer Look at Weak Label Learning
for Audio Events