This is both a comment for the paper and for the repo's README: the first python lib that came to my mind when I saw the review request was librosa. Could you - in the paper's statement of need - elaborate on why this lib is relevant in comparison to what librosa already provides? This would neatly fit with the other challengers (namely, Bob, SpeechPy and python_speech_features). The same argument could be made with the less well-known surfboard.

As a side note and quasi-request, it'd be nice to have such a comparison outlined in the README. If I were someone that came across this repo, i'd like to know how this is not "yet another audio feature extraction lib", and how this one might fit my needs (especially in comparison to other libs). If I were of bad faith, i'd say i'd like to see a bit of marketing :smiling_imp:

TypeError: dct(): incompatible function arguments. The following argument types are supported: 1. (a: array, type: int, axes: object = None, inorm: int = 0, out: object = None, nthreads: int = 1) -> array

segmentation fault occured when my program attempted to access the frames which returned by the function stride_trick the signal length is 1852416 in "stride_trick" a.size returns 3704832

Reference Issue

This is part of the code review for JOSS: https://github.com/openjournals/joss-reviews/issues/4739

This PR is based on the edits from #44 . That other PR should probably be merged before this one.

What does this implement/fix? Explain your changes.

This PR has has three aims:

• I'll add type hints to the function signatures (I hope you don't mind, but I really do love type hints, @hbredin might testify on this)
• Adding type hints sort of forces me to comb through the code and get a better understanding it of it
• If I see fishy things I might :
  • either add a # TODO or # WARNING comment for you to solve. You can probably solve them by directly pushing in this PR if you wish
  • correct them myself if they're extremely obvious. In this case you might want to check that I haven't done anything bad.

Reference Issue

This is a PR relating to https://github.com/openjournals/joss-reviews/issues/4739#issuecomment-1260768703

First of all, since this is our first contact, I have to say that this lib is very well made :smiley: , thanks a lot for your work! The code is very clean, the API is very clear, the tests are very exhaustive, and the documentation seems to be very complete.

The changes in this PR are somewhat nickpicky, but I think they really help making the lib more "standard".

What does this implement/fix? Explain your changes.

This fixes and "standardizes" a couple of things:

• I move most of the constants from the spafe/version.py to the setup.py , as this was not a standard thing to do. If this choice was very opinionated (i.e., you had a very good reason to do so), you can obviously revert that.
• I put the test and documentation dependencies in tests and docs , in the extras_require field. This is somewhat more standard, and allows the user to run pip install spafe[docs] or pip install spafe[tests] which I think is a bit cleaner. Note that running pip install spafe[anything] installs the "main" dependencies by default.
• I left the "main" dependencies in the requirements.txt, as i think this to be pretty practical.
• Having the ability to plot things is very good, but requiring the install of matplotlib by default is, I think, a bit bloated. I made this dependency optional, and added its install through pip install spafe[plotting]. You should probably say something about that in the install instructions.
• I relaxed the required versions for scipy/numpy ( >= instead of == , this is very useful if spafe has to share its env with some other libraries)
• I fixed the github actions in several ways:
  • I used a pip install instead of conda install, this makes the tests setup faster and much simpler
  • The old config didn't use the python-version matrix
  • I removed the tests for python 3.5 and 3.6 (they wouldn't be able to run anyway, the required numpy/scipy versions were not available for theses versions of python).
  • I added an action for the documentation. This couldn't run with the others actions, as you don't want a PR branch to activate a doc build.

By written in pycharm of the piece of the code in the document, I found that the "return" of the method is not a array of fbanks but a array of "a array of fbanks" and another array that I dont know its mean, I guess it is an array of center frequencies by print it out. Please fix up the document.

  File "/usr/local/lib/python3.9/site-packages/spafe/features/spfeats.py", line 279, in extract_feats
    feats["mode_frequency"] = frequencies[amplitudes.argmax()]
IndexError: index 3806427 is out of bounds for axis 0 with size 39168

  File "/usr/local/lib/python3.9/site-packages/spafe/features/spfeats.py", line 270, in extract_feats
    feats["peak_frequency"] = frequencies[np.argmax(amplitudes)]
IndexError: index 3806427 is out of bounds for axis 0 with size 39168

We are getting in the wrong way the argmax() of amplitudes when amplitudes dimensions is greater than 1.

Moreover, mode_frequency and peak_frequency have the same value, because amplitudes.argmax() and np.argmax(amplitudes) are the same, but this is a different issue https://github.com/SuperKogito/spafe/issues/34

I am working in audio files and I use this code to extract lfcc features:

audio_path = '03-01-02-02-01-01-01_norm.wav'

fs, sig = scipy.io.wavfile.read(audio_path)

lfcc = lfcc(sig, 13)

but I have this value error:

negative dimensions are not allowed

I think that it is because my sig array has negative values, can you give me a way to fix it? Thank you for yor help

pnccs = pncc(sig,fs,13) Traceback (most recent call last): File "", line 1, in File "D:\old data\Python\Python35\lib\site-packages\spafe\features\pncc.py", line 205, in pncc b=gammatone_filter.T) File "<array_function internals>", line 6, in dot ValueError: shapes (400,46) and (257,26) not aligned: 46 (dim 1) != 257 (dim 0)

i try this https://spafe.readthedocs.io/en/latest/features/mfcc.html, but got this "ImportError: cannot import name 'mfe' from 'spafe.features.mfcc'"

Reference Issue

closes https://github.com/SuperKogito/spafe/issues/14

What does this implement/fix? Explain your changes.

• Cast to int frame_length and frame_step as suggested from the issue.
• Remove unused variables

Any other comments?

Thanks https://github.com/aacarneiro/spafe/commit/95a8c785a7982db697a2ee5ff9f57845f5f56f5f @aacarneiro