search

bfs18 / nsynth_wavenet

parallel wavenet based on nsynth
106 stars 30 forks source link

KL loss #6

Closed maozhiqiang closed 6 years ago

maozhiqiang commented 6 years ago

the H_Ps should be used
screenshot from 2018-07-06 11-31-11 so H_Ps = tf.reduce_mean(log_scale) + 2 change to H_Ps = tf.reduce_mean(log_scale) + 2*T Is that right to understand

HallidayReadyOne commented 6 years ago

@maozhiqiang @bfs18 I am also a bit confused here, why not *T? Because E(∑1-T(ln s + 2)) ?