librosa.output.write_wav doesn't accept int16 typed arrays

Description

librosa.output.write_wav doesn’t accept int16 typed arrays, while it worked before. I found this is caused by #616, https://github.com/librosa/librosa/commit/f3114f280f5f9eaee5f2a69f90d3771b3d0eb5fa. Is this an intentional change or a bug?

Steps/Code to Reproduce

In [1]: import librosa; import numpy as np

In [2]: librosa.__version__
Out[2]: '0.5.1'

In [3]: x, sr = librosa.load(librosa.util.example_audio_file(), sr=None)

In [4]: librosa.output.write_wav("test.wav", (x * 32768).astype(np.int16), sr)
---------------------------------------------------------------------------
ParameterError                            Traceback (most recent call last)
<ipython-input-4-9736a6fc3adb> in <module>()
----> 1 librosa.output.write_wav("test.wav", (x * 32768).astype(np.int16), sr)

/home/ryuichi/Dropbox/python/librosa/librosa/output.py in write_wav(path, y, sr, norm)
    214 
    215     # Validate the buffer.  Stereo is okay here.
--> 216     util.valid_audio(y, mono=False)
    217 
    218     # normalize

/home/ryuichi/Dropbox/python/librosa/librosa/util/utils.py in valid_audio(y, mono)
    157 
    158     if not np.issubdtype(y.dtype, np.float):
--> 159         raise ParameterError('data must be floating-point')
    160 
    161     if mono and y.ndim != 1:

ParameterError: data must be floating-point

Expected Results

No errors.

Actual Results

As shown above.

Versions

In [5]: librosa.show_versions()
INSTALLED VERSIONS
------------------
python: 3.6.1 |Anaconda custom (64-bit)| (default, May 11 2017, 13:09:58) 
[GCC 4.4.7 20120313 (Red Hat 4.4.7-1)]

librosa: 0.5.1

audioread: installed, no version number available
numpy: 1.13.3
scipy: 0.19.1
sklearn: 0.19.1
joblib: 0.11
decorator: 4.0.11
six: 1.11.0
resampy: 0.2.0

/home/ryuichi/anaconda3/lib/python3.6/site-packages/sphinx/util/compat.py:40: RemovedInSphinx17Warning: sphinx.util.compat.Directive is deprecated and will be removed in Sphinx 1.7, please use docutils' instead.
  RemovedInSphinx17Warning)
numpydoc: installed, no version number available
sphinx: 1.6.3
sphinx_rtd_theme: 0.2.4
sphinxcontrib.versioning: 2.2.1
matplotlib: 2.0.2
numba: 0.35.0+10.g143f70e90

[stefan-balke]

I would keep it as is. If people need more flexibility they could use pysoundfile…maybe we should document that? 😃

[bmcfee]

Could this have changelog entry? This breaks some user code. e.g.,

Yes! Sorry for that – it’s linked in updated changelog for rc1, but I’ll make it more explicit for the final distribution. Note that doc build/deployment takes several hours, so you won’t see it right away in the main documentation.