librosa.util.normalize doesn't preserve dtype

In [1]: import librosa

In [2]: y, sr = librosa.load(librosa.util.example_audio_file())

In [3]: y.dtype
Out[3]: dtype('float32')

In [4]: y = librosa.util.normalize(y)

In [5]: y.dtype
Out[5]: dtype('float64')

[bmcfee]

Fixed by merging #636

[carlthome]

IMO dtype should be preserved, but not forced to be float32. It should be enough to just get rid of .astype(np.float) as np.abs will preserve dtype as expected AFAIK.

E.g.

>>> import numpy as np
>>> a = np.zeros(1, dtype=np.complex64)
>>> np.abs(a).dtype
dtype('float32')
>>> a = np.zeros(1, dtype=np.float64)
>>> np.abs(a).dtype
dtype('float64')
>>> a = np.zeros(1, dtype=np.float32)
>>> np.abs(a).dtype
dtype('float32')