What?

A tool to extract audio files from video files -- for use with the Music Thing Modular Radio Music. https://github.com/TomWhitwell/RadioMusic

Feed it a file full of video files, and it will extract the audios into wav format and distribute them (in order) into a set of directories, suitable for copying to an SD card.

Setup

Prerequsites

• python - I used 2.7.13
• ffmpeg - must be in $PATH
• some video files you want to extract audio from
• bunch of disk space
• recent Radio Music firmware that supports .wav files

Input file format

Prepare an input text file that tells the extractilizer what videos you want to extract audio from. The format is one video file per line. You can optionally specify an audio stream number and a colon at the beginning of the file. It might look like this:

/home/videos/nuclear_tests.mkv
/home/videos/my_birthday.avi
4:/home/videos/big_buck_bunny.iso
/home/video/performance_art.ogg

Note that the 3rd file specifies the audio stream 3 from the dvd, which might be a specific language that you want.

Usage

usage: extractilizer.py [-h] [-m MINUTES] [-o OUTDIR] infile

positional arguments:
  infile                the input file containing video files, one per line

optional arguments:
  -h, --help            show this help message and exit
  -m MINUTES, --minutes MINUTES
                        minutes per split output file (default = 10)
  -o OUTDIR, --outdir OUTDIR
                        output dir (default=output)

Example

$ ./extractilizer.py -m 15 -o o2 videos.txt

How it works

For each file in the input file:

• Extract the audio stream to a file
  • Remove all silence below -50dB for 0.5 seconds
• Normalize audio by amplifying the peak to 0dB
• Split the normalized file into chunks of N minutes (default=10, or commandline arg)
• Distribute the files, in order, among 16 numbered output dirs
  • "Extra" files are loaded into the last dir

Limitations/improvements

• Can only extract one audio stream per video file.
• Does it also work with audio files as input? Not sure, but would/could be useful.
• Parameters around silence detection are hard-coded, should make parametric
• Crashes hard -- could be made more flexible/recoverable/resumable
• Volume normalization is very basic, and increases the noise floor? Maybe some users would like one of the more sophisticated/dynamic normalizers?
• Make number of banks a param, in case user doesn't want to make all 16?

Please fork me and make me better!

Did you try it out? Let me know how it went: [email protected]