I've learned a little more about a few popular sound codecs recently. One item, mentioned in the last article, is that the MD ATRAC encoding has gotten better over the years.

Of course, another thing I found, after encoding gigs of stuff, and you probably know this already, is that 160 kbps is much safer for encoding your stuff than 128 kbps. Yes, it's true that 128 kbps works most of the time, but you usually find out it didn't work too late - when you're listening to it on an airplane, and then it's too late to re-encode it.

Of course, mp3.com encodes everything at 128 kbps to keep downloads manageable.

Which brings me to my main point: Microsoft claims that their WMA encoder sounds great at 64 kbps. My tests using MusicMatch jukebox were that the sound wasn't any good at all! MP3 wasn't too hot either, but it didn't make horrible screeching sounds the way WMA did! What's this about?

Amusingly, Microsoft, at their demo site, used the Dire Strait's Money for Nothing as a test. This was the same song I had a lot of trouble with! (Here's the link - hopefully it will hang around for awhile: you be the judge. [Okay, technically this isn't a Microsoft site - but Microsoft pointed me to it as "independent testing". But who put the test people up to the idea of downsampling first?])

Down at the bottom of the page I discovered a key fact:

"Note on Methodology: All compressed files were downsampled using Sonic Foundry's Sound Forge. No other pre-processing was performed."

What's this? So each file is downsampled before it is encoded!?

I gotta tell you: this makes all the difference in the world! It also redefines what "near-CD" quality means! If near-CD quality means 22 kHz then, well, I just don't know what!

For one thing, if you are willing to expand your definition of near-CD quality to mean half the quality of a CD then that opens up a lot of options!

Downsampling before encoding … wow!

But here's my question: if this is such a great thing to do (and it is - it eliminates all the frequency aliasing you get otherwise), then why isn't it built into the encoder? Huh? If I select 64 kbps as an encoding rate, then the encoder should downsample the frequency for me! Don't ya think?

Well, this really turned my head around. After all, I'm the one that said listening to 11K 8-bit mono on a jet airplane isn't so bad. And it was downsampled by Cool Edit. So I'm not against doing this!

It's just that if you innocently set your encoding rate to 64 kbps that you are going to get crap!

So, really, it should be built into the encoder. Right? Right.

Also, if you're happy with 22 kHz playback, then another encoding option is to encode using ADPCM that's already been sampled down to 22K. This gives you files half again as big as 44 kHz playback and sounds pretty good. I saw a title recently that did this. I may fool around with that a bit some time.

In the meantime, maybe we can get Fraunhofer and Microsoft to build the downsampling code into their encoders so we can get decent sounding encodings at low sample rates without having to first sample into a PCM file and then downsample it and then covert it. Besides the fact that most people don't have the tools to do this, as I said in another article moving that much data around is a pain and somewhat defeats the need for compression!

Well, it's probably just a matter of time. No doubt, the primary reasons it's not done at this point are (1) no one thought of it, and (2) it would slow the whole process down. Although option (1) begs the question: if all of this stuff is encoded in the frequency domain, then how hard can it be to skim off the upper frequencies as part of the encoding process?

It's all very strange if you ask me. But the short lesson is this: for home and personal use, you're probably screwed. But if you want to use WMA in a game, be sure to downsample first. This should be no problem because you or your sound guy should already have the tools to do it.