librosa becomes slower the longer it runs

Description

My code is basically:

for fn in lots_of_flac_files:
  audio, sample_rate = librosa.load(fn, sr=None)
  mfccs = librosa.feature.mfcc(
    audio, sr=sample_rate,
    n_mfcc=40,
    hop_length=int(0.01 * sample_rate), n_fft=int(0.025 * sample_rate))

All files are about the same length. I added some simple code to estimate the remaining time. The strange thing is that this estimation did not went down, but went up only in the beginning, i.e. it became slower and slower. How could that be? Also, I noticed that the memory consumption seems to slowly increase all the time.

Expected Results

Every call to load and mfcc takes about the same time.

Actual Results

The time to execute load and mfcc always increases. (Maybe only one of them; I didn’t tested.)

Versions

>>> import platform; print(platform.platform())
Linux-4.9.0-3-amd64-x86_64-with-debian-9.3
>>> import sys; print("Python", sys.version)
Python 3.5.3 (default, Jan 19 2017, 14:11:04) 
[GCC 6.3.0 20170118]
>>> import numpy; print("NumPy", numpy.__version__)
NumPy 1.14.1
>>> import scipy; print("SciPy", scipy.__version__)
SciPy 1.0.0
>>> import librosa; print("librosa", librosa.__version__)
librosa 0.6.0
>>> 

[albertz]

Well, not totally resolved, but we can say that it is not a librosa bug, so yes, this can be closed.

[bmcfee]

Why don’t you use that by default?

Because libsndfile does not support mp3 yet. When it does, we’ll switch over.