Then you go to copy the file to your iPod and notice that iTunes thinks it is 36 seconds long. And you deleted everything but the one big file. Moron.

Here’s how to recover from that:

1. Download faad2 and neroAacEnc.

2. Open a command line console (cmd.exe), and set it to a low priority with the Task Manager.

3. Run something like this:

   faad.exe -o - "inputfile.m4b" | neroAacEnc.exe -if - -of "outputfile.m4a" -br 64000 -lc -ignorelength

4. Re-tag your file with MP4Box or neroAacTag or AtomicParsley.

If you want to add chapter marks, even if they are just at spaced-out intervals:

1. Download MP4Box from the GPAC project.

2. Create a text file:

   CHAPTER1=00:00:00.00
   CHAPTER1NAME=Hour 1
   CHAPTER2=01:00:00.00
   CHAPTER2NAME=Hour 2
   CHAPTER3=02:00:00.00
   CHAPTER3NAME=Hour 3

   Etc, etc.

3. Run something like:

   MP4Box.exe -chap chapters.txt -ipod file.m4a

Anecdotally, I've found that while the chapter tagging of MP4Box works well, the other metadata tagging isn't always seen by iTunes or foobar2000. I've had to run another set of tags via neroAacTag, and those seem to work. It's convoluted, but still easier than trying to juggle a bunch of hour-long files.

Tools:

• ffmpeg or faac: Convert from WAV/MP3 to AAC format, then wrap in M4A/M4B MPEG-4 container.
• madplay: Stream MP3 to WAV for faac, if necessary
• tag: Remove ID3 tags.
• vbrfix: Doctor up conjoined MP3 files to fix timing issues.
• mp3wrap: Conjoin MP3 files.

I’m pretty sure that everything on that list works just as well on Mac and Linux.

You do not need both ffmpeg and faac, as ffmpeg has faac built-in. Personally, I think using faac offers more control, but whatever. If you are going to use ffmpeg, then you can probably get away with this:

[code lang="winbatch"]ffmpeg -i infile.mp3 -sameq outfile.m4a[/code]

That seemed to work, but I honestly didn’t look too hard. Instead, I prefer doing it the more explicit and verbose way with faac:

[code lang="winbatch"]madplay -m -o wave:- infile.mp3 | faac -b 48 -o outfile.m4b --no-midside -w --artist Artist --title Title --genre Audiobook --album Album --disc "8/13" --year 2009 --cover-art cover.jpg -[/code]

Note the hyphen at the end! It’s there for a reason!

Of course, if you need to conjoin MP3 files into one big file, you need to do that first:

[code lang="winbatch"]mp3wrap -v outfile.mp3 infiles*.mp3
tag --remove outfile.mp3
vbrfix fromfile.mp3 tofile.mp3[/code]

You can also do this the hard way, with copy /b but you must strip out the tags before you do it:

[code lang="winbatch"]tag --remove *.mp3
copy /b in01.mp3+in02.mp3 joined.mp3
vbrfix joined.mp3 out.mp3[/code]

If your files are all constant bitrate, then the vbrfix step isn’t necessary … but it won’t try to fix anything if it doesn’t have to, so it’s probably a good habit to get into out of paranoia anyway.

I haven’t figure out how to include chapter/disc breaks into the M4B file yet. That’s my next research project.

Interestingly enough, my last.fm profile paints a different picture:

Yes it was prepared entirely with ColdFusion. I’ve been mulling over doing some data mining visualization with CF, so this was sort of a dry run for that. In a nutshell:

iTunesMusicLibrary.cfc

Read in the iTunes Music Library.xml file that iTunes provides for interoperability. Convert it into a giant query of tracks and their play count and most recent play date. Get a list of the top 80 most-played albums, where the album play count is the maximum of its tracks’ play counts.

LastFmUser.cfc

Fetch my weekly play chart via the Last.fm API and convert it to a query. Also fetch cover art.

AmazonCoverArt.cfc

Query and fetch the cover art from Amazon’s XML web service, assuming we couldn’t find it on Last.fm.

TreemapList.cfc

Do some basic treemapping work to figure out how big each album cover should be, relative to all of the others.

SquarifiedTreemap.cfc

Do some more advanced decision-making based on trying to keep the album covers as square as possible. This component decides exactly where each cover fits in the image. Based on the idea by J.J. van Wijk and, loosely, on the Java code by Martin Wattenberg and Ben Bederson.

TreemapRenderer.cfc

First draft generated plain HTML with absolute CSS layout. Second draft used cfimage to make a single composite image.

Along the way I made heavy use of WDDX to checkpoint my progress and keep me from slamming Amazon’s web service, or having to re-parse my 5MB iTunes XML over and over again. It’s about 900 lines of code in total, probably triple that once you count in code that I refactored and deleted.

A fun exercise, to be sure, but also very frustrating. I forget that Query-of-Queries is still not much more than a toy when it comes to doing anything truly useful. I want to believe that it can do even the basic stuff, but it chokes all to easily. Several hours of development were spent working around QoQ bugs.

And don’t get me started on the sad state of affairs that is album cover art.

I must say that I’m pleased with the results of the cfimage portion. It’s still too slow to do lots of image compositing for a busy site, but for non-sustained use it’s workable.

I’ve been considering doing something like this as a visualizer for Banshee. For that version, though, I’d want to have an Invert button that inverted the play counts for each, highlighting music that you haven’t listened to in a while. If I work on my treemap-fu, I might even get it to where it will group all of an artists albums next to each other. (Though I have doubts that this would look as nice.)