[SoundStage!]Fringe with Greg Smith
Back Issue Article
November 1999

MP3: Compression Culture Collision

Hobbyists love to polarize themselves into opposing groups. Audiophiles are certainly no exception. Analog vs. digital, solid state vs. tubes, it's hard to escape falling into a camp and shouting at the unbelievers on the other side. Right now, there's a new audio fault line cracking open. Last time there was a disturbance in this area, audiophiles landed on the winning side. We might not be so lucky again.

When you want to get sensory data into a computer, typically audio or video, it needs to make a transition into digital form first. Once there, such data takes up an enormous amount of memory. A single minute of audio, recorded at CD quality, takes up approximately 10MB; video is even greater. Computers never have enough memory or hard-drive space, so since the computer's inception, programmers have used compression techniques to reduce the amount of space information takes up. Compression comes in two forms. Formats like the popular Zip are lossless; whatever you put in one side of the compression machine comes out the other side identical to the source. The reduction in size from this type of compression varies according to the type of data being compressed. For example, a typical Windows program might compress anywhere from 40-60% of its original size when "zipped." Data files, with lots of redundant bytes, can shrink an order of magnitude or more. CD-quality audio doesn't compress very well for two reasons. First, the 16-bit sample rate doesn't mesh well with a typical compression program, which look at 8 bits worth of data at a time. There are audio-specific compression schemes available for your PC that address this problem; I'm searching for one worth recommending right now. But even these programs, and similar schemes like the Meridian Lossless Packing popularized in the DVD-Audio specification, generally achieve about a 50% reduction in size when working on non-trivial audio. You might get an 80% reduction compressing simplistic material from the top-40 chart, but there's no guarantee. Graphics on the Internet are universally compressed into formats like GIF and JPEG, which take advantage of a couple of tricks to reduce file sizes. Since there are often long stretches of identical data patterns, these can be represented in an alternate format. For example, a repeating texture can be stored once and reproduced many times, without storing the full texture in each place. That technique is again vulnerable to encountering runs of data that look random to the compression scheme, resulting in no net savings in size.

Lossy compression is a different sort of animal. Human senses have a limited resolution at which they can operate, and this can be exploited to reduce the amount of data required to produce an "identical" copy of something. The JPEG format works with an adjustable "quality" parameter; as the required quality drops, the file reduces in size. Compressing graphics results in subtle losses in fine detail and color resolution. Video formats like MPEG take this a step further, by also looking at the differences between successive frames and compressing that. A typical movie scene contains a static background and moving bodies on top. A smart motion compression scheme stores the static elements once, then encodes the changes to generate a new frame. Only when the background has changed massively is a new key frame stored. Again, there is an adjustable quality factor here during compression where aggressive attempts to reduce size give motion artifacts, blockiness, and other problems video reviewers love to point out.

The MPEG group didn't just work on video. Recently, their MPEG 1 Layer 3 audio encoding scheme, generally abbreviated to MP3, has become extremely popular. MP3 compression technology has been successfully applied on two fronts. It's possible to take a regular audio CD, place it in your computer's CD-ROM, and perform digital audio extraction (DAE) on the audio, converting it into a computer-friendly format like the WAV file. This process, popularly known as "ripping," is substantially error-free when done with a CD-ROM optimized for DAE use. Plextor makes the best units on the market for that purpose, available only in SCSI, while Panasonic models seem to be widely recommended for the IDE bus. Anyway, once a CD track is on your computer's hard drive, taking about 10MB of space per minute, it can be manipulated with standard computer audio tools. It's also straightforward to then compress the file. Converting to the MP3 format typically results in a file 10% or less in size, which means a three-minute-long single goes from 30MB down to less than 3MB. If you crank the numbers a bit, that means a 6GB hard drive, average nowadays on new systems, could easily hold 100 hours of compressed MP3 audio. That's around 75 full-length albums, more than most people have in their CD collections.

And that's assuming a fairly high quality. As with all lossy compression, we again find a variable factor: the target data rate. People who care about fidelity usually aim for a target rate of 128Kb/s or more. (Note the switch in abbreviation here: KB or MB are kilo-bytes and mega-bytes, while Kb and Mb refer to kilo-bits and mega-bits. Divide bits by 8 to convert to bytes.) If you're willing to sacrifice a (excuse the pun) bit of sound, you can easily pump the data rate down to even 16Kb/s or less. This brings us to the second application of MP3. Since nobody wants to wait a long time to be entertained if they can avoid it, streaming audio protocols were designed. If you have a 33.6Kb/s modem, you can certainly decode a 16Kb/s MP3 audio stream as fast as it is being transferred. While players generally pause for a bit when they first get started to buffer the audio a bit and even out the sound in case of momentary data disruptions, once you get going, an MP3 file or stream can be played back in real time, with you listening to it as its transferred. While some commercial products, like the RealNetworks player, have similar streaming audio capabilities, this comes with some proprietary technology restrictions that the MP3 crowd isn't as restrained by (although there are patents on some of the MP3 encoding technology that still apply).

So just about anybody can take their music collection, convert it to MP3, and listen to it on their computer with totally free software. If you have a CD recorder, you can record over 12 hours worth of MP3 audio where only 74 minutes of uncompressed sound will fit. There are sites all over the Internet that have archives of music to download, some legal, some not. Forward-thinking record companies like Ryko even sell music in MP3 form; download one file for a dollar or so, and you've got a copy of a song you like, without needing to buy the whole album or settling only for the official singles. And anywhere there are MP3 files, if the streaming data rate is slower than your Internet connection, you may be able to hear the music while it downloads. This capability has helped spawn Internet radio stations. Broadcasters "multicast" their audio to many computers at once, so once you tune into a station you can hear the music being played over the network.

MP3 is unquestionably fun and still remains a bit on the cutting edge side of things, although with the introduction of products like Diamond's Rio (a standalone MP3 player), the technology is going very mainstream. People ask me on a regular basis why I don't write more about Internet audio in general and MP3 in specific. There's a very good reason. MP3 playback is about as directly opposed to high-end audio as it is possible to be, and I find it of very limited utility in its current form.

Early digital systems suffered from many flaws, including poor mastering, bad D/A conversion, inadequate filtering, and a host of other problems you can point out. It's just in the last few years that good-sounding digital products have appeared at just about every price point. It's been a while since I heard a CD player that sounded really bad. Computer sound, on the other hand, still has quite a ways to go. Typical sound cards are still quite harsh, tinny, and noisy relative to high-end audio standards. Recently I've started to track some very affordable cards with reasonable digital outputs on them, even in the $40 price range. This is the direction I'm clearly heading towards. A digital out helps isolate the noisy interior of my computer from injecting garbage into the sound I hear. Simultaneously, computer DVD-ROM drives have gained market share, and many of those include Dolby Digital AC-3 outputs. Right now I'm upgrading my computer sound setup to use all digital outs and be capable of decoding surround sound.

And you expect me to be interested in MP3? CD audio runs at 1376Kb/s. Dolby Digital puts five channels into, normally, 448Kb/s. Those kind of figures are just barely good enough to produce digital audio I consider acceptable. The new 24-bit/96KHz audio DVD discs rolling out push the bit rate up even higher. That's the future of really great audio; that's where I want to go. 128Kb/s MP3? I don't think so. While it's possible to encode my own music with higher bits rates, I already own those CDs. Most of the stuff I've been downloading from the Internet runs at 128Kb/s or slower. To support that speed, those files are often sampled at 22kHz instead of the CD standard 44kHz. This means you only get a frequency response up to 11kHz. Near CD quality? That's not even FM radio quality. The last time I listened to an audio format with specs that poor, the songs were split across 8 tracks.

And the legality of the whole scene disturbs me. I know a lot of college students who use MP3 technology extensively. Since their music-buying budget is so limited, and the popular music they listen to so often disposable, I can understand their reliance on the format. Understand, though, that while it has plenty of other applications, MP3 has gained its popularly by being an easily accessible way to pirate music recordings. Ultimately recording artists and possibly the record industry at large will distribute music digitally, but this is not what the majority of the current usage involves. Internet radio stations that are actually legit, paying the appropriate licensing fees so that the artists who wrote the music are compensated for their work, tend to have extremely short play lists. Yes, the music industry charges quite a bit for their products, but using that as an excuse to never buy them and just pirate songs instead is juvenile. You don't have to listen to music if you don't like the terms it is being distributed under.

Fans of MP3 tout the technology as "good enough" and often get caught in a typical audio vicious circle. Since the computer sound card doesn't sound very good, they get cheap speakers to go with it. And since the speakers are poor, it's difficult to notice the flaws in the compression technology. So why bother worrying about having the best possible source? The system-building approach that high-end audio uses, where every components matters and each should be as optimized as possible, is obviously the result of a very different listening culture. Accepting any single component of the reproduction chain as flawed, but good enough despite better alternatives, is a dangerous precedent.

Compressed music hasn't had much commercial success in the past. The Philips DCC format died despite backward compatibility with the cassette tape, the most popular form of music sales at the time. Sony's Minidisc has only recently started to gain market acceptance in the US. Outcry from audiophiles about the problems with sound quality certainly didn't help sales of those formats. Despite that history, it looks like MP3 and similar computer-based compression technology is well along a sharp climb in adoption. I'm all for downloading samples of music with this sort of program. Lately I've been finding a whole bunch of long lost songs using the excerpts CDNOW provides to identify music I hadn't heard in a while. But when it comes time to listen seriously, I buy the full CDs or even pay a premium for remastered versions. Let's hope that the rush to adopt the latest in computer/audio convergence doesn't accidentally hurt industry efforts to provide the best audio possible. It would sure be a shame to have the reproduction quality of the music we listen to take such a big step backwards after so much progress.

...Greg Smith
gregs@soundstage.com

[SoundStage!]All Contents
Copyright © 1999 SoundStage!
All Rights Reserved