Lots of python libraries for reading mp3 files require spawning new processes that execute cli utilities. For example, AudioSegment runs ffpeg underneath. This is suboptimal for my particular use case because I convert files and I write to a database which works better with in memory files.
From reading torchaudio source code I could not get a straightforward answer since it uses different backends and C++ wrapper code.
Could you please confirm torchaudio.load does/does not spawn a new process for loading mp3?