Issues with evaluation (std and salience for beat/downbeat)

[in-conscious]

Hi! I was trying to realize the evaluation part of the code according to the paper (see the attached picture) Pop Music Transformer: Beat-based Modeling and Generation of Expressive Pop Piano Compositions:

However, There must be some mistakes with my calculation because of the perceptable distincts between my result of beat std and downbeat salience and the one the paper. Also, I found difficulty in realizing the part of downbeat std because of the different description (perhaps?) of the paper and the file for madmom. The data presented in my result is based on the train set (Remi/data/train) and madmom, and I think I transferred the .midi files into correct .wav files:

I've viewed https://madmom.readthedocs.io/en/latest/modules/features/downbeats.html for related information but still failed to solve it by myself. Therefore, could you please share the evaluation part with me? Thank you so much!

[remyhuang]

please attach your code, thanks.

[in-conscious]

Here's my code, thanks. `

def Eval_rhythm(audio_path):

  print("path ", audio_path)
  # RNNDownBeatProcessor
  act = downbeats.RNNDownBeatProcessor()(audio_path)       
  # DBNDownBeatTrackingProcessor
  proc2 = downbeats.DBNDownBeatTrackingProcessor(beats_per_bar=[4, 4], fps=100)
  result = proc2(act)
  # print("act = ", act)
  # print("---" * 50)
  # print(result)

  # Beat Std
  rawbeat_std = [i[0] for i in result]
  beat_new = [0] * len(rawbeat_std)
  beat_new[0] = rawbeat_std[0]
  for i in range(1, len(rawbeat_std)):
      beat_new[i] = rawbeat_std[i] - rawbeat_std[i - 1]
  beat_std = np.std(beat_new, ddof=1)
  print("Beat Std = ", beat_std)

  # Downbeat Std
  rawdownbeat_std = [i[1] for i in result]
  downbeat_new = [0] * len(rawdownbeat_std)
  downbeat_new[0] = rawdownbeat_std[0]
  for i in range(1, len(rawdownbeat_std)):
      downbeat_new[i] = rawdownbeat_std[i] - rawdownbeat_std[i - 1]
  downbeat_std = np.std(downbeat_new, ddof=1)
  print("Downbeat Std = ", downbeat_std)

  # Downbeat Salience
  downbeat_salience = [i[1] for i in act]
  salience = sum(downbeat_salience) / len(downbeat_salience)
  print("Downbeat Salience = ", salience)
  
  return beat_std, downbeat_std, salience

`

[remyhuang]

Take a quick look at your code.

1. why beat_new[0] = rawbeat_std[0], but beat_new[i] = rawbeat_std[i] - rawbeat_std[i - 1]
2. is "downbeat" salience, not "all" salience

[in-conscious]

I guess τBt-τB(t-1) for τB0 should be (τB0 - 0)? That's because I failed to find the code of the referred model. It was missing according to the link in the paper of that author (for the several edition I've found). I also wrote to the website master of the institute he was working in for the new link of that code but there's still no reply.

I think I was calculating the downbeat salience. There's a nonstandard in the name of it ("salience") and I'll revise it. Thanks again!

[in-conscious]

Would you like to share the evaluation code?

[MarcusLee-awsome]

Can you share the code of the evaluation。I can't find it in the website