On my system, it was using Gstreamer, which had several issues (e.g. #57, #62 and #63).
Then I tried PySoundFile, which was even faster (0:04:31 vs 0:05:38), and does not have any of those issues.
A related discussion for librosa is here.
So, I think it would make sense for audioread to add PySoundFile as an additional backend, and even make it the default.
On my system, it was using Gstreamer, which had several issues (e.g. #57, #62 and #63). Then I tried PySoundFile, which was even faster (0:04:31 vs 0:05:38), and does not have any of those issues. A related discussion for librosa is here. So, I think it would make sense for audioread to add PySoundFile as an additional backend, and even make it the default.