End ground truth boundaries is falsely doubled

[georgid]

Then boundaries_id = 'gt' the boundaries are converted first to frame indices here. Then once the labeling algorithm outputs the results this line recomputes the boundaries from the frame indices, potentially adding an end-of-audio boundary. If however the end-of-audio boundary is present in the ground truth boundaries, it becomes doubled after this line is executed. This is due to some discrepancy in the conversion to frame indices and back to time.

Here is an example

est_times of ground truth input boundaries:  [0, 12.689583333333333, 27.68958333333333, 42.18958333333333, 56.380208333333336, 70.63020833333333, 83.38020833333333, 98.88020833333333, 125.88020833333333, 145.33020833333333, 157]

est_indices:
[   0  274  597  909 1215 1521 1796 2130 2711 3130 3381]

est_times:
[  0.          12.72453515  27.72462585  42.21387755  56.4244898
  70.63510204  83.4060771   98.9170068  125.8985941  145.3569161
 157.01333333 157.058322  ] 

here the config:

<class 'dict'>: {'annot_beats': False, 'feature': 'pcp', 'framesync': True, 'dirichlet': False, 'xmeans': False, 'k': 4, '2dfmc_offset': 4, 'label_norm_feats': inf, 'label_norm_floor': 0.1, 'label_norm_min_db': -80, 'features': <msaf.features.PCP object at 0x124eba8d0>, 'hier': False}

[urinieto]

Thanks for reporting this. I will try to work on a workaround soon. This seems indeed problematic and should be fixed asap.

[georgid]

Hey @urinieto I did some workaround and am trying to push a separate branch, but get this error message

georgid$ git push --set-upstream origin workaround_issue_104
remote: Permission to urinieto/msaf.git denied to georgid.
fatal: unable to access 'https://github.com/urinieto/msaf.git/': The requested URL returned error: 403```

[urinieto]

Hey Georgi, please fork it and then do the PR. Thanks so much!