Loading fragments of file is slow in newer version

Describe the bug Loading a piece of a longer audio file is much slower in v0.9.2 than it was in v0.8.1. For instance, loading a 2-hour wav file takes 440 ms, and loading the first 5 seconds using librosa 0.8.1 only takes about 1 ms. However, in the 0.9.2 package version, loading the first 5 seconds is taking about 130 ms.

Apologies if this is a duplicate issue, I have not found a similar one when searching the issues.

To Reproduce

import librosa
from time import time

t=time()
librosa.load('two_hour_file.wav') 
print(time()-t) # 440 ms, good

t=time()
librosa.load('two_hour_file.wav',offset=0,duration=2)
print(time()-t) # 130 ms, slow! 

Expected behavior Loading a short piece of audio should be very fast, as it was in previous package versions.

Software versions

Linux-5.15.0-43-generic-x86_64-with-glibc2.17
Python 3.8.13 (default, Mar 28 2022, 11:38:47)
[GCC 7.5.0]
NumPy 1.22.0
SciPy 1.9.0
librosa 0.9.2
INSTALLED VERSIONS
------------------
python: 3.8.13 (default, Mar 28 2022, 11:38:47)
[GCC 7.5.0]

librosa: 0.9.2

audioread: 2.1.9
numpy: 1.22.0
scipy: 1.9.0
sklearn: 1.1.2
joblib: 1.1.0
decorator: 5.1.1
soundfile: 0.10.3
resampy: 0.3.1
numba: 0.56.0
pooch: v1.6.0
packaging: 21.3

numpydoc: None
sphinx: None
sphinx_rtd_theme: None
sphinx_multiversion: None
sphinx_gallery: None
mir_eval: None
ipython: None
sphinxcontrib-svg2pdfconverter: None
pytest: None
pytest-mpl: None
pytest-cov: None
matplotlib: 3.5.2
samplerate: None
soxr: None
contextlib2: installed, no version number available
presets: None

[agkphysics]

Yes I can confirm that with audioread==3.0.0 the audio loading times are the fast for both librosa==0.9.2 and librosa==0.9.1.

[bmcfee]

https://github.com/beetbox/audioread/pull/126 is merged and should fix this issue for you, if you’re willing to run on a development version of audioread. scratch that, audioread 3.0 is out, so no dev install necessary! Let me know if this resolves things for you and we can close this issue out.