Can't read mp3 from BytesIO: File contains data in an unknown format.

Describe the bug An MP3 file that can be loaded just fine from disk fails to be loaded when offered as an io.BytesIO object.

If this is a known restriction, it could be documented better.

To Reproduce

import librosa
import io

# This works fine.
print(librosa.load('/tmp/test.mp3'))

# This doesn't.
with open('/tmp/test.mp3', 'rb') as f:
    data = f.read()
print(librosa.load(io.BytesIO(data)))

Actual output

/.../.venv/lib/python3.8/site-packages/librosa/core/audio.py:162: UserWarning: PySoundFile failed. Trying audioread instead.
  warnings.warn("PySoundFile failed. Trying audioread instead.")
(array([0., 0., 0., ..., 0., 0., 0.], dtype=float32), 22050)
Traceback (most recent call last):
  File "/tmp/test.py", line 10, in <module>
    print(librosa.load(io.BytesIO(data)))
  File "/.../.venv/lib/python3.8/site-packages/librosa/core/audio.py", line 165, in load
    raise (exc)
  File "/.../.venv/lib/python3.8/site-packages/librosa/core/audio.py", line 146, in load
    with sf.SoundFile(path) as sf_desc:
  File "/.../.venv/lib/python3.8/site-packages/soundfile.py", line 629, in __init__
    self._file = self._open(file, mode_int, closefd)
  File "/.../.venv/lib/python3.8/site-packages/soundfile.py", line 1183, in _open
    _error_check(_snd.sf_error(file_ptr),
  File "/.../.venv/lib/python3.8/site-packages/soundfile.py", line 1357, in _error_check
    raise RuntimeError(prefix + _ffi.string(err_str).decode('utf-8', 'replace'))
RuntimeError: Error opening <_io.BytesIO object at 0x7ff3086254f0>: File contains data in an unknown format.

Software versions

Linux-5.8.14-arch1-1-x86_64-with-glibc2.2.5
Python 3.8.6 (default, Sep 30 2020, 04:00:38) 
[GCC 10.2.0]
NumPy 1.18.2
SciPy 1.5.4
librosa 0.8.0
INSTALLED VERSIONS
------------------
python: 3.8.6 (default, Sep 30 2020, 04:00:38) 
[GCC 10.2.0]

librosa: 0.8.0

audioread: 2.1.9
numpy: 1.18.2
scipy: 1.5.4
sklearn: 0.23.2
joblib: 0.17.0
decorator: 4.4.2
soundfile: 0.10.3
resampy: 0.2.2
numba: 0.51.2

numpydoc: None
sphinx: None
sphinx_rtd_theme: None
sphinxcontrib.versioning: None
sphinx-gallery: None
pytest: None
pytest-mpl: None
pytest-cov: None
matplotlib: 3.2.1
presets: None

[bmcfee]

The problem here is that mp3 is not supported by soundfile (yet), so librosa will fall back to audioread. Audioread doesn’t generally support BytesIO input (because it in turn wraps several possible decoder backends, like ffmpeg, gstreamer, etc).

I’d be open to expanding the documentation here, but don’t have the bandwidth to handle it myself right now.

[ttencate]

@xuboot See my earlier comment.