I am trying to benchmark some popular python audio libs for audio work in python. I intend to test how the Read/Write of these libs are. I tested with wav files with length 1, 2, 5, 10, 30, 60, 300, 600, 1800 with the following libs: PyTorch Audio, librosa, soundfile, pydub. The speed is shown in the graph below. Can anyone explain me why it is so fast and why pydub is the fastest? I am also wondering how big software like Meta or Youtube handle such small audio files in large quantities. I'm trying to understand it if we take away from Python, it's a slow language. P.S. I understand that Pytorch is slow because of the overhead of converting to Tensor. Continue reading...
