I'm doing research for a project that involves thousands of hours of radio communications downloaded from Broadcastify archives, which are divided into 30 minute mp3s. To make it easier on myself I wrote a script to remove the silence and concatenate the mp3s into a single output file. So each day's audio goes into its own directory, and then the script spits out a clean, short file for the whole day. Unfortunately, that script sucked because it only worked on one file at a time and would have taken hundreds of hours to process everything.
Enter concurrent.futures and the ProcessPoolExecutor class. It's a way around Python's Global Interpreter Lock to allow for CPU-bound parallelism. So, on the low end 4 core laptop I'm doing all this stuff on, it defaults to four workers instead of just the one I was using before. It also makes my computer hum.
This script uses pydub to manipulate audio files.
For Arch you can get it from the AUR or otherwise pip install pydub.
By default min_silence_len=5000 is set to 5000ms (5 seconds) but you can tune it to whatever floats your boat.
python deadair.py <input_directory> <output_file.mp3>
Depending on the file size, quantity and amount of processing required this can still take a while.