allow dtype specification in abs2, rms

[bmcfee]

Reference Issue

fixes #1656

What does this implement/fix? Explain your changes.

This PR allows a user to specify a target dtype in abs2 and rms calculations.

This resolves the issue with rms calculation on integer inputs noted in #1656 . For example, if dtype is not specified, we get overflow:

In [14]: x = np.arange(1, 32768, dtype=np.int16)

In [15]: librosa.util.abs2(x)
Out[15]: array([1, 4, 9, ..., 9, 4, 1], dtype=int16)

If dtype is specified, we get proper outputs:

In [16]: librosa.util.abs2(x, dtype=np.float32)
Out[16]: 
array([1.00000000e+00, 4.00000000e+00, 9.00000000e+00, ...,
       1.07354522e+09, 1.07361075e+09, 1.07367629e+09], dtype=float32)

Float32 is the default for rms now, as the result will be averaged and then sqrt'ed anyway.

Any other comments?

I've also snuck in a replacement of np.abs(S)**2 by util.abs2(S) in the rms calculation. Not sure how we missed that one before, but it's fixed now.

[codecov]

Codecov Report

Base: 98.79% // Head: 98.79% // Increases project coverage by +0.00% 🎉

Coverage data is based on head (3916583) compared to base (26439ac).
Patch coverage: 100.00% of modified lines in pull request are covered.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #1665   +/-   ##
=======================================
  Coverage   98.79%   98.79%           
=======================================
  Files          34       34           
  Lines        4559     4563    +4     
=======================================
+ Hits         4504     4508    +4     
  Misses         55       55           

Flag	Coverage Δ
unittests	98.79% <100.00%> (+<0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
librosa/feature/spectral.py	99.17% <100.00%> (+<0.01%)	⬆️
librosa/util/utils.py	100.00% <100.00%> (ø)

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

☔ View full report at Codecov.
📢 Do you have feedback about the report comment? Let us know in this issue.