Chord Recognition Using Librosa-problem with librosa.feature.chroma_cqt

[maneesha dendukuri]

Hey everyone,

 So, I'm trying to implement chord recognition using a very simple model of template matching. But I tend to get errors in using the Librosa.feature.chroma_cqt() function.

I found this GitHub repo, which does the same, and i still get the same errors when i followed this repo.Link to Repo

Please help.

the error log is here:

File "PycharmProjects/pyML/chordgram.py", line 146, in <module>

    C = chromagram(file)

  File "\pyML\chromagram1.py", line 24, in chromagram

    chromagram = librosa.feature.chroma_cqt(y=y_harmonic, sr=sr, hop_length=hop_length)

  File "\Anaconda3\lib\site-packages\librosa\feature\spectral.py", line 936, in chroma_cqt

    real=False))

  File "Anaconda3\lib\site-packages\librosa\core\constantq.py", line 230, in cqt

    sparsity)

  File "\Anaconda3\lib\site-packages\librosa\core\constantq.py", line 490, in __cqt_filter_fft

    pad_fft=True)

  File "\Anaconda3\lib\site-packages\librosa\filters.py", line 601, in constant_q

    sig = sig * __float_window(window)(ilen)

  File "\Anaconda3\lib\site-packages\librosa\filters.py", line 447, in _wrap

    window = window_function(n, *args, **kwargs)

  File "\Anaconda3\lib\site-packages\scipy\signal\windows.py", line 807, in hann

    if _len_guards(M):

  File "\Anaconda3\lib\site-packages\scipy\signal\windows.py", line 19, in _len_guards

    raise ValueError('Window length M must be a non-negative integer')

ValueError: Window length M must be a non-negative integer

[Brian McFee]

What versions of librosa, numpy, and scipy do you have?

Decoding the error: it looks like the problem is that the CQT basis is trying to build a filter with a fractional length (in terms of samples).  This used to work in older versions of scipy, but in late 2016 they added the length guards.

We fixed this a while back in librosa, see this issue: https://github.com/librosa/librosa/issues/449  and the fix: https://github.com/librosa/librosa/pull/452 (merged 2016-12-01), so if you're on at least librosa 0.5 it shouldn't be a problem.

[maneesha dendukuri]

Thank you so much.It worked.Updating librosa to 0.6 totally fixed it.(I was on 0.4 before).