Label Preprocessing Problem

[Sodayeyeye]

Hello! I would like to know how you deal with the silence tag in the tag, because I see that there is no silence in your tag

[tae-jun]

Hi, start and end marks silences at the beginning and the end :)

[Sodayeyeye]

Thank you for your reply. Is the silence tag still converted to silence during preprocessing? When I was training, This will give you an error “ File "/ opt/anaconda3 / envs/allin/lib/python3.9 / site - packages/allin1 / training/data/eventconverters/eventconverters py", line 153, in labels = np.array([self.label_map[l] for l in labels]) KeyError: 'silence',” how do you handle this

[tae-jun]

I excluded silence from segment labels. For example, the original tags for 0004_abc.txt is:

0.0 silence
2.666656 verse
28.300543 chorus
58.26318 verse
78.843287 chorus
106.457736 bridge
138.973914 chorus
153.598533 chorus
168.288673 chorus
180.490951 end

and here is the post-processed one:

0.046349206 verse
25.947641723    chorus
56.01276644 verse
76.656689342    chorus
104.331972789   bridge
136.907755102   chorus
151.570068027   chorus
166.298412698   chorus
175.830204082   end

And please refer to this comment for other conversion rule: https://github.com/mir-aidj/all-in-one/issues/8#issuecomment-1798473615

Thanks!

[Sodayeyeye]

Thank you very much for your reply! May I ask you what the specific steps are? Did we do this for audio and structure labels and beat labels before training? Because I see that after you do this, the duration of the audio changes, and the beats label should change accordingly. Thank you for your patient reply!