We face some problems when we do manual annotation. Should we label
Also, another question is, how can we handle the no-vocal part between two sentences? Is it music with singing or pure-music? Would it influence our training results?
- music/non-music?
- singing/non-singing?
- music with singing/music without sining/no music
- singing with music/singing without music/no singing
- ...... something else
Also, another question is, how can we handle the no-vocal part between two sentences? Is it music with singing or pure-music? Would it influence our training results?
2 則留言:
we handle all these questions in the way as below:
1 pure music
2 music with singing
3 ambiguous
4 nonmusic
and if two singing sequences are separated too far, then we separate them with a pure music sequence, or a ambiguous one, which depends on cases.
so you are doing multiple-class discrimination. You can simply use the "One-against-all" approach as I mentioned in the lecture. SVM or Adaboost either one is OK for building such discriminative models.
張貼留言