Walk a directory of .mp3, .wav, or .m4a files, and transcode each file with LAME, putting the new file into a cloned directory. For .mp3 and .wav files, lame is used; for .m4a files, faad is used. The tool will also copy images (.jpg, .png, .pdf) into the new directory tree.

I like to listen to music when I ride my bike, run, etc., so I keep a fair amount of music on my phone. I got a new mobile phone without an SD card slot :(, so I have precious little storage on the new phone. On my old phone, I had about 30 GB of music on an SD card, but I didn't want to remove a bunch of songs from my collection.

This tool ameliorates the (too widespread) problem of not having an SD card to expand the music storage of your mobile phone. With lame_walker.py, you can transcode your music files to .mp3 files that use a lower average variable bitrate. When I'm running, biking, etc., I don't need super high-fidelity audio (I listen to a lot of black metal on cheap earbuds!), so we can pretty safely use a lower bitrate, which results in significantly smaller .mp3 files. Thus, I can store more songs per unit of storage on my phone.

The prototypical call to lame to transcode an input .mp3 to a lower bitrate is something like

lame --abr 160 -b 96 input.mp3 output.mp3

We can also use one of their presets:

lame --preset medium input.mp3 output.mp3

Or use the variable bit rate quality parameterization:

lame -V 7 input.mp3 output.mp3

lame_walker.py is simply a wrapper around lame (and faad) to walk an input directory and call the transcoder in multiple processes using Python's multiprocessing module. Using the default --lame-arg (-V 7), I transcoded a 20 GB directory tree to about 15 GB.

Instead of using lame_walker.py, transcoding files in a directory could also be done with find and xargs. The following one-liner will transcode the .mp3s in a directory in place using 16 processes.

find inoutdir -type f -name "*.mp3" -print0 | xargs -0r -P16 -n1 -I % bash -c 'lame -V 7 --quiet "%" "%.tmp" && mv "%.tmp" "%"'

But lame_walker.py shows a nice progress bar. (It used to show bitrate histograms for the multiple workers in a curses display. This was neat, but only marginally useful so I've rewritten the code with simplicity in mind.)

• lame for (.mp3, .wav) -> .mp3 transcoding.
• faad for .m4a -> .wav transcoding (we then transcode the .wav to .mp3 with lame)
• We use Python 3.6+ and the associated standard library

On Arch Linux, you can install lame and faad with pacman -S lame faad.

To use the tool, put the files you want to transcode in a directory, say original, for this example. Let's say original looks like this:

original
+-- artist_1
|   +-- album_1
|   |   +-- track_1.mp3
|   |   +-- track_2.mp3
|   |   +-- cover.jpg
|   +-- album_2
|   |   +-- track_1.m4a
|   |   +-- track_2.m4a
|   |   +-- album_sleeve.pdf
+-- artist_2
|   +-- album_1
|   |   +-- track_1.mp3

To walk original and transcode the music files and copy images, run

python3 /path/to/lame_walker.py original/ resampled/

The argparse help is

usage: lame_walker.py [-h] [--num-workers NUM_WORKERS] [--lame-args LAME_ARGS] indir outdir

Convert MP3 files from a directory tree to use average/variable bitrate and copy the transcoded files to a cloned directory tree.

positional arguments:
  indir                 The directory of original MP3 files.
  outdir                The directory of output MP3 files.

optional arguments:
  -h, --help            show this help message and exit
  --num-workers NUM_WORKERS
                        The number of worker processes to run simultaneously.
  --lame-args LAME_ARGS
                        The optional arguments pased to `lame`.