unable to install dcase_util from source due to latest librosa

[rjkat]

Description

The latest librosa depends on llvmlite, which is broken with LLVM 6.0.0, which means it's not possible to import dcase_util when installing from source. This is necessary when working on a platform which doesn't have precompiled binaries. Please consider downgrading librosa.

LLVM issue references: https://bugs.llvm.org/show_bug.cgi?id=35947 https://reviews.llvm.org/D44140

Steps/Code to Reproduce

Install dcase_util from source:

pip install -v --ignore-installed --no-binary :all: dcase_util

import dcase_util

Expected Results

Should be able to import dcase_util.

Actual Results

$ python
Python 2.7.11 (default, May 25 2016, 14:24:28)
[GCC 4.4.7 20120313 (Red Hat 4.4.7-16)] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> import dcase_util
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/short/wa66/rk9125/lib/python2.7/site-packages/dcase_util/__init__.py", line 6, in <module>
    from . import containers
  File "/short/wa66/rk9125/lib/python2.7/site-packages/dcase_util/containers/__init__.py", line 948, in <module>
    from .audio import *
  File "/short/wa66/rk9125/lib/python2.7/site-packages/dcase_util/containers/audio.py", line 11, in <module>
    import librosa
  File "/short/wa66/rk9125/lib/python2.7/site-packages/librosa/__init__.py", line 12, in <module>
    from . import core
  File "/short/wa66/rk9125/lib/python2.7/site-packages/librosa/core/__init__.py", line 102, in <module>
    from .time_frequency import *  # pylint: disable=wildcard-import
  File "/short/wa66/rk9125/lib/python2.7/site-packages/librosa/core/time_frequency.py", line 10, in <module>
    from ..util.exceptions import ParameterError
  File "/short/wa66/rk9125/lib/python2.7/site-packages/librosa/util/__init__.py", line 70, in <module>
    from . import decorators
  File "/short/wa66/rk9125/lib/python2.7/site-packages/librosa/util/decorators.py", line 9, in <module>
    from numba.decorators import jit as optional_jit
  File "/short/wa66/rk9125/lib/python2.7/site-packages/numba/__init__.py", line 10, in <module>
    from . import config, errors, runtests, types
  File "/short/wa66/rk9125/lib/python2.7/site-packages/numba/config.py", line 11, in <module>
    import llvmlite.binding as ll
  File "/short/wa66/rk9125/lib/python2.7/site-packages/llvmlite/binding/__init__.py", line 6, in <module>
    from .dylib import *
  File "/short/wa66/rk9125/lib/python2.7/site-packages/llvmlite/binding/dylib.py", line 4, in <module>
    from . import ffi
  File "/short/wa66/rk9125/lib/python2.7/site-packages/llvmlite/binding/ffi.py", line 128, in <module>
    raise e
OSError: /short/wa66/rk9125/lib/python2.7/site-packages/llvmlite/binding/libllvmlite.so: undefined symbol: LLVMInitializeInstCombine

Versions

NumPy 1.11.0 SciPy 1.1.0 Matplotlib 1.5.1 librosa 0.6.0

[toni-heittola]

Librosa is used in the case_util for audio file reading and feature extraction related functionality. dcase_util should work with librosa version 0.5.1 in most use cases. There are only a few cases where 0.6.0 is required, e.g. RMSEnergyExtractor parameter n_fft would not work. However, I do not see any strong reason to downgrade librosa version requirement. I would not like to maintain different code/functionality depending on librosa version.

You could try to downgrade your librosa version to 0.5.1, and see can you import dcase_util then.

[rjkat]

Downgrading to 0.5.1 didn't help, I ended up having to go all the way back to 0.4.

librosa 0.4 doesn't have support for resizing user-supplied windows for the stft. I encountered this error:

librosa.util.exceptions.ParameterError: Size mismatch between n_fft and len(window)

I ended up changing n_fft to 640 to make it match the length of the window. I imagine this will cause my feature extraction to be slower.

Is this an appropriate way of resolving this issue? Do you have any other advice?

[toni-heittola]

Your extracted features might not match with the ones extracted with current librosa version. However, if you are not trying to recreate some system implemented on top of dcase_util this should not be a problem.

On which platform your are operating? Would using anaconda/miniconda type of python distribution help?

[toni-heittola]

This issue was closed as there was no follow up from author for at least 4 weeks. Feel free to reopen issue if needed.