search

andabi / music-source-separation

Deep neural networks for separating singing voice from music written in TensorFlow
795 stars 150 forks source link

Some confusing about "bss eval metric" #11

Open ShengleiH opened 6 years ago

ShengleiH commented 6 years ago

Hello, in your evaluation code, there are two 'gnsdr' scores, one for music, and the other for vocal. 

# Write the score of BSS metrics
 tf.summary.scalar('GNSDR_music', gnsdr[0])
tf.summary.scalar('GSIR_music', gsir[0])
tf.summary.scalar('GSAR_music', gsar[0])
tf.summary.scalar('GNSDR_vocal', gnsdr[1])
tf.summary.scalar('GSIR_vocal', gsir[1])
tf.summary.scalar('GSAR_vocal', gsar[1])

But in Po-sen's paper, he only use one gnsdr, I am wondering, which one he used? The music's gnsdr, or the vocal's gnsdr?

Thanks a lot!!! :)