Audio Codec Quality Shootout

http://www.extremetech.com/print_article2/0,1217,a=123493,00.asp

Audio Codec Quality Shootout April 2, 2004

EDITOR RATING: By Jason Cross

Millions of people convert their CD collections to a compressed digital audio format, often without considering the format or settings they use. With several different popular formats and lots of quality settings, many users just pick an application they like and use whatever the default settings are. ExtremeTech readers -- generally a bit more hands-on and do-it-yourself -- are more likely to seek out other applications or players and encode their audio exactly how they want. Certainly a top consideration should be compatibility. We all have different audio devices, and it doesn't matter how good your music sounds if it won't play on your hardware. We'll leave that decision up to you; you know what software and devices you have and what they're capable of playing. Encoding time is another possible consideration, but honestly, modern computers encode audio so quickly that the difference in speed between one format and another is practically a non-issue. This article focuses exclusively on sound quality. There have been audio quality comparison tests before, but they usually focus on a single pair of codecs – often comparing the codec du jour against MP3. These are usually conducted in special listening labs, where extremely high-end audio equipment is used in a very controlled environment with expert, well-trained listeners. This method has its merits, but we think there's value it doing it a different way for a change. We wanted to find out which codec and settings sound best to normal people under normal listening conditions – at home on their own stereos and computers, using their own speakers or headphones. If you need to get up to speed on digital audio terms and formats, check out our recent Digital Audio Primer. There are literally dozens of audio codecs out there now, but only a few are relevant to the average consumer. Our audio quality comparisons will focus on the four most popular formats: MP3, AAC, WMA, and Ogg Vorbis: MP3: By far the most ubiquitous format for compressed digital audio, MP3 needs little introduction. Though distributing encoders and decoders requires a license, there are scores of free players and encoders

1 of 14

10/3/2006 16:05

Audio Codec Quality Shootout

http://www.extremetech.com/print_article2/0,1217,a=123493,00.asp

available. There's quite a range of quality amongst the different encoders, so we used what we felt was the best-sounding one you can get. LAME isn't technically an encoder (it has to be compiled into one) but a quick Google search will turn up plenty of already-compiled versions. They can be combined with free CD rippers like Exact Audio Copy to produce an all-in-one ripping package. LAME is considerably slower than other MP3 encoders, but produces the best audio quality. The LAME build we used was 3.95.1. Windows Media Audio 9 Another familiar codec, this one is built into the latest Windows Media Player (among other applications) and offers a wealth of features. The standard stereo version is most common, but there are versions that offer 24bit/96KHz sound at up to 5.1 channels, and even a version that offers mathematically lossless compression. Our tests focused on the common stereo version, but we threw in a Lossless track just to see if listeners could pick it out. We used the free Windows Media Encoder to encode our test clips. AAC AAC has been around for a long time as part of the MPEG-2 spec, but has been updated for better low-bitrate quality in the MPEG-4 spec. It wasn't very popular among home users until the explosive popularity of Apple's iTunes application, which defaults to AAC encoding for ripping CDs. As it is by far the most popular home AAC encoding application now, we used iTunes to encode our AAC test files. Not many portable players support AAC, but the iPod does, and it's the current market leader. Ogg Vorbis Ogg Vorbis is similar to MP3 or AAC compression formats, but with one important difference: It's completely free, unpatented, and open-source. Its royalty-free nature has made it popular with some game developers and software publishers -- it's used in several top PC games including Unreal Tournament 2003, Serious Sam: The Second Encounter, and Harry Potter and the Chamber of Secrets. You can find out more about Ogg Vorbis and download the source or tools from Vorbis.com. Only a handful of portable players support .ogg files, but many are capable of it and, since it's free, vendors can add support if their user base demands it. It's not enough to simply test the four most popular compressed audio formats, though. There's always the matter of what bitrate to choose. We tested each clip Digital Audio Primer at three common bitrates: 64k per second (useful for smaller flash iTunes Bad, WMA memory-based devices), 128k (for larger flash memory devices or small hard Good drive based players), and a quality-based variable bitrate setting with the quality up to 98% (for large hard drive based players or storage on your PC). The exception here is AAC. Though true variable bitrate support is possible in the AAC standard, the most popular AAC encoder, iTunes, only offers constant bitrate options at this time. Therefore, our AAC tests did not include the VBR setting. It is critically important to understand our test methodology, because we wanted to make this a human test, not a lab test. It is by no accident that we tried to get our music out of the sterile, perfectly reproducible lab and into the real world – controlled lab experiments have value, but we wanted to test the quality of these audio codecs in the environments and using the equipment that regular people would use. To do this, we had to devise a single-blind test method for listeners to rate music using their

2 of 14

10/3/2006 16:05

Audio Codec Quality Shootout

http://www.extremetech.com/print_article2/0,1217,a=123493,00.asp

own home or PC audio equipment, yet not know which codec they were listening to. We first created four audio clips, each one chosen not necessarily because the music is "good," but because it represents the typical musical attributes of a particular genre. These clips also have attributes that present different challenges to audio compression schemes. We took these four tracks and encoded them into each of the four codecs described on the previous page, at bitrates of 64k, 128k, and a 98% quality variable bitrate. We then converted these compressed audio clips to standard redbook CD audio tracks. This introduces no additional artifacts and perfectly preserves the sound properties of the various compression formats and bitrates – we confirmed this through spectrum analysis, listening tests, and by speaking with audio compression experts. These were burned to a set of CDs – one for each of the four songs – with the original source audio as track 1 and all the other tracks randomized. Listeners were to listen to these tracks on their best home audio equipment using quality speakers or headphones and then rate each track from 1 to 5, with 5 being a perfect reproduction of the source track (as far as they could tell). Note that this was a blind test; the listeners had no idea which track represented which codec or bitrate, and each CD had a different random track order. We threw in a WMA Lossless track on each CD too, just to see if all the listeners would give it a 5 rating. When the scores were returned, we took the geometric mean of the submitted scores to produce an overall quality rating from 1 to 5. Our sample set included half a dozen listeners, some with very well trained ears and expensive audio equipment, and others that are just music lovers with a good set of headphones. We did augment our "out of the lab" tests with a relatively simple "in the lab" test. Using Sound Forge 7, we created a spectrum analysis graph of each compression format and bitrate for all four songs, which we'll use to compare how much sound data is preserved at various frequencies. Note that you cannot simply rely on a spectrum analysis graph to tell you which codec sounds best. These codecs all function by throwing away sound information that would be least audible, so sound quality is not just a function of what's missing from the spectrum, but also of the quality of what's there. The spectrum analysis only shows the amount of energy at different frequencies – it does not show whether that frequency is signal or noise, and shouldn't be taken as the sole measurement of a codec's capabilities. It's best to focus on the results of the real-world listening tests; the spectrum analysis graphs are presented simply for your edification. The four CDs, representing four musical styles, contained the following tracks: CD1 – Jazz: Squirrel Nut Zippers, "Got My Own Thing Now" This lively tune in the 30's flapper / Dixieland jazz style is a good test of encoding quality, with everything from a walking slap-bass line to persistent hi-hat and ride cymbals, and a little brassy sax and trumpet thrown in for good measure. The heavy layering of instruments and constant cymbals are the biggest challenges to an audio codec here. CD2 – Techno: Fluke, "Atom Bomb" Meant to represent the typical encoding load of techno and electronic music, this contains mostly synthesized instrumentation with extreme highs and lows, a good amount of layering, and typical techno-style post-processing effects. This should be the easiest of the four clips to compress – the

3 of 14

10/3/2006 16:05

Audio Codec Quality Shootout

http://www.extremetech.com/print_article2/0,1217,a=123493,00.asp

predictable sound waves of the artificially-created sounds in techno music don't tend to give audio codecs a very hard time. CD3 – Pop: No Doubt, "Just a Girl" Pop music is a broad category, so we picked a song that represents the general instrumentation, recording setup, and layering of most pop music. There's bass and lead guitar work with various levels of distortion, a drum set that doesn't let up on the kick drum or hi-hat, and a decent pop vocalist. Like most pop music, this song transitions from a relatively calm verse to a more raucous chorus. Accurately reproducing the guitar distortion with all the other "stuff" going on, including constant hi-hat crashing, should be the hardest thing for the codecs to deal with. CD4 – Classical: James Horner, "Elora Danan (Theme to Willow)" This piece from the main theme to the movie Willow is a good representation of typical large-ensemble classical music, which is generally very hard to reproduce well. There's plenty of dynamic range here, with large string ensembles, ambient concert hall reverb, a bit of brass work, rolling tympani, and large cymbal crashes. The lead melody is played by a shakuhachi, a Japanese wood flute that is very "breathy" and has a lot of character, making it hard to recreate well. Classical music is an incredibly broad category, and while it may be tempting to use a Segovia guitar sonata or Chopin nocturne, these are generally easy for modern day codecs to deal with, and we went for something a little more layered and intense to give the compressors more of a workout. All of our compression work took place on a typical mid-range desktop PC with the following configuration, although the quality of the resulting files should be unaffected by the machine used: CPU

Pentium 4C 2.4GHz CPU with an 800MHz FSB and Hyper-Threading

Motherboard Chipset

Intel 875PBZ

RAM

1 GB Kingston HyperX DDR400 SDRAM

Hard Drive

Western Digital WD2000JB 200GB

Video

ATI Radeon 9800 XT

Audio

Sound Blaster Audigy 2

OS

Windows XP Professional with Service Pack 1

Let's take a look at how well the various codecs performed on the first audio track, made to represent the encoding load of lively jazz music. First, let's look at the spectrum analysis. Here's what the original track looks like: And here's the track compressed with Windows Media Audio 9's Lossless codec. Although the compression ratio here is only about 2:1, and therefore only really useful for archiving or for those with really big hard drives, it does create a perfectly mathematically lossless recreation of the source material. For the remaining lossy compressed tracks, let's start at the bottom and look at the 64k files. In

4 of 14

10/3/2006 16:05

Audio Codec Quality Shootout

http://www.extremetech.com/print_article2/0,1217,a=123493,00.asp

order to achieve such high compression, the codecs perform what is called a "low-pass filter," effectively eliminating all sound above a certain frequency, since those are hardest to hear anyway. This gives the codec less data to work with. Note that the MP3 codec has eliminated everything over around 9KHz, while the falloff for WMA9 and AAC is closer to 13KHz. Ogg fares best here, with a very sharp falloff at a higher frequency of around 15KHz. When we step up the bitrate to 128k, perhaps the most commonly used bitrate, we notice that things have changed considerably. The MP3 and AAC files both have the worst low-pass filter rolloffs, cutting out right around 16KHz, although the AAC track is much closer to the source material's spectrum in the middle frequencies. WMA9 fares a little better, matching the source well in the low and middle frequencies, starting to roll off around 16KHz, and ultimately giving up the ghost at nearly 18KHz. Ogg Vorbis performs best in this particular analysis, smoothly tapering off around 20KHz. Looking at the variable bitrate files, only the MP3 really has a problem, with the low-pass filter kicking in around 17KHz. That's pretty high, but WMA9 keeps trucking to around 20KHz, although it's clearly losing some audio signal between 18 and 20KHz. Ogg again fares best, smoothly tapering off beyond 20KHz. Most adults can't hear frequencies that high, anyway. Of course, the spectrum analysis isn't nearly as important as our subject's listening tests. After all, you don't look at a graph of your music, you listen to it. Here are the scores of the various codecs and bitrates, along with the best and worst rating they received. 64k Encoding Codec

Geometric Mean

Best

Worst

MP3

1.64

3

1

WMA9

3.81

5

2

Ogg Vorbis

3.98

5

2

AAC

3.17

5

1

Codec

Geometric Mean

Best

Worst

MP3

3.95

5

3

WMA9

4.13

5

3

Ogg Vorbis

3.81

5

2

AAC

3.81

5

2

128k Encoding

Variable Bitrate Encoding

5 of 14

10/3/2006 16:05

Audio Codec Quality Shootout

http://www.extremetech.com/print_article2/0,1217,a=123493,00.asp

Codec

Geometric Mean

Best

Worst

MP3

3.17

5

1

WMA9

3.95

5

3

Ogg Vorbis

3.13

5

1

Clearly, MP3 has a really hard time at a bitrate of only 64k. If you have a flash memory-based portable music player and want to squeeze on more music by encoding your files with a bitrate of 64k, you'll want to avoid MP3. Ogg Vorbis performs best at those low bitrates, slightly edging out WMA9, while WMA9 has the best scores at 128k and with variable bitrate encoding. It's especially interesting to see that all three codecs had lower scores in the VBR tests than they did when encoded at a constant 128k bitrate, even though the resulting VBR files were often 2-4 times the size. This is due mostly to low scores from just one or two listeners, while the rest scored them very highly. It is unsurprising that the 128k scores were more consistent, but we still would have expected the VBR scores to be higher overall. Moving on to the techno music track, we expect to see similar spectrum analysis graphs, but much higher listening scores. The synthetic nature of techno produces what could only be described as "well-formed sound waves" that are easier for compressors to deal with. Here's the spectrum analysis of the original track. The WMA9 Lossless compressed track looks absolutely identical, as you would expect from a lossless codec. At a constant bitrate of only 64kilobits per second, the graphs look eerily similar to the 64k graphs of the Jazz track. Once again Ogg Vorbis has the best-looking graph, followed by AAC, then WMA9, with MP3 trailing far behind with a horrible drop-off around 9KHz. The MP3 track seems to cut off at a bit lower frequency on the Techno track than it did on the Jazz track. The low-pass filter seems to behave the same on the other three tracks, however. Notice that both Ogg Vorbis and WMA9 start to deviate from the original tracks spectrum graph around 15KHz, losing some fidelity before the low-pass filter kicks in. The VBR tracks again look pretty good for WMA9 and Ogg Vorbis, while the MP3 low-pass eliminates almost everything above 16KHz. Moving on to the all-important listening tests, here's how our subjects scored the codecs at various bitrates. 64k Encoding Codec

6 of 14

Geometric Mean

Best

Worst

10/3/2006 16:05

Audio Codec Quality Shootout

http://www.extremetech.com/print_article2/0,1217,a=123493,00.asp

MP3

1.74

4

1

WMA9

3.98

5

2

Ogg Vorbis

4.13

5

3

AAC

4.13

5

3

Codec

Geometric Mean

Best

Worst

MP3

4.00

4

4

WMA9

4.51

5

3

Ogg Vorbis

3.78

4

3

AAC

4.37

5

4

Codec

Geometric Mean

Best

Worst

MP3

3.73

5

3

WMA9

3.64

4

2

Ogg Vorbis

4.13

5

3

128k Encoding

Variable Bitrate Encoding

Again we see MP3 totally flounder at the low bitrate of 64k, while Ogg Vorbis and AAC perform best. WMA9's score is nothing to sneeze at, though. At 128k, WMA9 does extremely well with a score over 4.5, and Ogg Vorbis fares surprisingly poor, losing out to even the venerable MP3 format. Again we see all three VBR codecs score lower in variable bitrate mode than at a constant 128k bitrate. Will the trend continue with other codecs? There is generally a lot of distortion present in modern pop and rock music, and plenty of post-processing done in the studio. This makes it difficult for listeners to determine if audio artifacts are part of the noise that's supposed to be there, or if it was introduced by the compression. Still, the heavy layering and cymbals should give the codecs a good workout. The original source CD audio track has a spectrum graph that looks like this. Once again, the WMA9 Lossless track looks absolutely identical. The pattern repeats itself. Since these spectrum graphs show most clearly the effects of the low-pass filter algorithms, the various codecs appear to stack up the same each time. Ogg Vorbis clearly has the least aggressive low-pass filter, while MP3 has the most aggressive. At the 64k bitrate, this is very important, but it becomes less of an issue at higher bitrates where the low-pass

7 of 14

10/3/2006 16:05

Audio Codec Quality Shootout

http://www.extremetech.com/print_article2/0,1217,a=123493,00.asp

filter eliminates much less of the sound spectrum. By the time WMA9 and Ogg Vorbis start cutting out sound frequency, it's starting to push the limits of human hearing. Both start to lose fidelity around 16KHz, but Ogg Vorbis tapers off smoothly, reproducing at least some audio in the 18-20KHz range. MP3 and AAC cut off sharply at a disappointing 16KHz. MP3's low-pass filter still kicks in at 16KHz, while WMA9 and Ogg Vorbis stay true all the way up to 20KHz. Looking at spectrum graphs is nice, but what really matters is how good the various compression schemes sound. Here's how the Pop track scored in the listening tests. 64k Encoding Codec

Geometric Mean

Best

Worst

MP3

1.15

2

1

WMA9

2.89

5

2

Ogg Vorbis

2.27

4

1

AAC

2.76

5

1

Codec

Geometric Mean

Best

Worst

MP3

3.25

5

2

WMA9

4.13

5

3

Ogg Vorbis

4.32

5

3

AAC

4.57

5

4

Codec

Geometric Mean

Best

Worst

MP3

4.57

5

4

WMA9

4.32

5

3

Ogg Vorbis

4.51

5

3

128k Encoding

Variable Bitrate Encoding

This clip seems to be the hardest one, so far, to encode at 64k per second, with none of the codecs producing a score over 3. This is a good case for making sure you don't "blanket-encode" all your music at 64k if you can help it. WMA9 did best, but none of the scores are worth bragging about. At

8 of 14

10/3/2006 16:05

Audio Codec Quality Shootout

http://www.extremetech.com/print_article2/0,1217,a=123493,00.asp

128k things change considerably, with AAC performing best and MP3 bringing up the rear with a score less that 4. This is also the first track where the variable bitrate scores are higher than the 128k constant-bitrate scores, with both MP3 and Ogg Vorbis delivering fantastic performance. Classical music varies from simple guitar solos to grandiose orchestras with choral accompaniment, and everything in between. Our song choice includes some fairly typical orchestral material – low rumbling strings, rolling tympani, grand cymbal crashes, and sharp brass accents – but the lead melody is performed by a breathy wooden Japanese flute called a shakuhachi, performed by Kazu Matsui. The complex characteristics of this instrument should be especially hard for the codecs to recreate well. Let's take a look at the original track's spectrum analysis graph. The spectrum analysis is a bit different than the previous tracks, but nonetheless, WMA9 Lossless reproduces it perfectly. The compression ratio for this lossless track was slightly less than 2:1, but such is the price of perfect audio reproduction. This 64k graph is a bit more interesting than the others. While the low-pass filter cutoffs happen at the same places, it's very interesting to see that the compressed tracks start to deviate pretty seriously from the original track, even at as low as 4-6KHz.The compressed tracks all have a far more wavy line than the original, meaning that at some frequencies, sound information is just getting lost. The excessively wavy spectrum graph at mid-range frequencies all but disappears at the higher bitrate of 128k, but both MP3 and AAC still roll off at 16KHz, while the low-pass filters for Ogg Vorbis and WMA9 are more forgiving. The graphs for variable bitrate compression certainly look the most like the original, including the spike up around 19KHz that was missing at other bitrates. Simply looking at the graphs, we would expect the 64k encoding to turn out terribly, the 128k to sound all right but not ideal, and the variable bitrate encoded clips to be best of all. Did this pan out in the blind listening tests? 64k Encoding Codec

Geometric Mean

Best

Worst

MP3

1.32

2

1

WMA9

2.72

5

1

Ogg Vorbis

3.13

5

2

AAC

1.32

2

1

128k Encoding

9 of 14

10/3/2006 16:05

Audio Codec Quality Shootout

http://www.extremetech.com/print_article2/0,1217,a=123493,00.asp

Codec

Geometric Mean

Best

Worst

MP3

4.18

5

4

WMA9

4.32

5

3

Ogg Vorbis

3.90

5

3

AAC

4.37

5

3

Codec

Geometric Mean

Best

Worst

MP3

3.95

5

3

WMA9

4.57

5

4

Ogg Vorbis

3.90

5

3

Variable Bitrate Encoding

The low bitrate 64k files sound pretty bad, but at least Ogg Vorbis manages a score above 3. It scores lowest at 128k, however, while WMA9 and AAC are essentially tied. The variable bitrate scores are sort of odd – MP3 scored worse than it did at 128k, Ogg Vorbis the same, and WMA9 did far better with a fantastic score of 4.57. Now that we've seen the breakdown for each of the four different test tracks, let's take a look at how the codecs performed overall. The geometric mean for the ratings of each codec, across all four music tracks, is as follows: 64k Encoding Codec

Geometric Mean

MP3

1.44

WMA9

3.15

Ogg Vorbis

3.29

AAC

2.97

At the space-saving bitrate of 64 kilobits per second, Ogg Vorbis clearly performs best. Our spectrum analysis graphs clearly show that the low-pass filter employed by all these codecs is very aggressive at this low bitrate in order to give the codec less data to try to squeeze into such a small amount of room. The Ogg Vorbis low-pass filter is set to a much higher frequency at 64 kilobits, and this no doubt plays an important role in its quality. WMA9 performed admirably, scoring over 3. MP3 simply shouldn't be used at all if possible – even the best encoder we could find didn't manage total overall score over 1.5. Since none of the codecs could get to a score of 3.5, let alone the desirable cutoff of 4.0, you should really not compress your music at a bitrate of 64k unless you really need to

10 of 14

10/3/2006 16:05

Audio Codec Quality Shootout

http://www.extremetech.com/print_article2/0,1217,a=123493,00.asp

fit a lot of audio in a small amount of memory. 128k Encoding Codec

Geometric Mean

MP3

4.06

WMA9

4.27

Ogg Vorbis

3.94

AAC

4.27

128 kilobits per second is not exactly a space-eating rate in today's world of hard drive based music players and 200 gigabyte desktop hard drives, but it's twice the rate of those 64k samples and that really makes all the difference. MP3 and Ogg Vorbis scored nearly the same, right around 4, which amounts to pretty good sound reproduction. Our experience has shown that not every MP3 encoder produces the high-quality results LAME 3.95.1 does, so it's probable that the built-in encoders in programs like MusicMatch or iTunes wouldn't fare so well. iTunes AAC encoder ends up matching WMA9 at this bitrate, and their scores look very good. VBR Encoding Codec

Geometric Mean

MP3

3.82

WMA9

4.10

Ogg Vorbis

3.88

WMA9 Lossless

4.52

Variable bitrate encoding is typically seen as a good way to get the best sound quality in a given file size. It's great for portable players or PCs, but not very good for streaming over the Internet, where the variance in the amount of data that is needed from one moment to the next plays havoc on net connections. What's especially interesting here is that the total overall scores for VBR encoders are lower than the scores for a constant bitrate of 128k, even though we set our encoders to target a 98% quality rate and the resulting files were often 2-4 times the size of the 128k files. Also, the low-pass filters are much less aggressive with these high VBR settings. So why the lower scores? The VBR settings may have scored lower because they are actually more accurate than the 128k versions. Most of the noise in a recording resides at very high frequencies. The low-pass filter that eliminates high frequency sound at 128k may be removing some of the subtle nuances of a recording, but it also eliminates most of the inherent noise. The listeners, not knowing what codecs or bitrates they were listening to, may have thought the VBR encodings sounded worse than the

11 of 14

10/3/2006 16:05

Audio Codec Quality Shootout

http://www.extremetech.com/print_article2/0,1217,a=123493,00.asp

128k samples because they more accurately reflect the noise in the source material. Notice that we also included a WMA9 Lossless score on the VBR table. We included one such track on each CD, purely out of curiosity. We are able to prove that this codec, which only achieves a compression ratio of about 2:1 on most musical tracks, is indeed mathematically lossless. It sounds exactly like the source material and produces an identical spectrum analysis graph, without fail. This would lead us to believe that the only score it can receive is a 5.0, right? But when our listeners didn't know what they were listening to, some of them scored it lower. Psychologists call this the "Pygmalion Effect," or self-fulfilling prophecy. Our listeners expected to hear compressed audio tracks and for there to be artifacts, and therefore couldn't bring themselves (on the whole) the give the lossless compression track a perfect score. Even though it was a perfect reproduction, they scored it as less than such because that is what they expected to hear. Before we get around to rating these four codecs, we should reiterate exactly what this article was meant to test. The high degree of control made possible by conducting listening tests on specialized, never-changing hardware in a lab has its merits, but that's been done before. Our goal was to have normal people listen to various audio codecs using the same hardware and in the same environments in which they actually listen to music. Our test group consisted of musicians with well-trained ears, plain old music lovers, and even a young child. The audio equipment they all used was pretty good – no chintzy earbud headphones or $40 PC speakers – and varied from studio monitor headphones to pricy home theater equipment. This was designed to be a real-world listening test, not a lab test. Bear that in mind when you look at our overall scores. While we have focused on sound quality here, there are several other important considerations when choosing which audio codec to encode your music with. Of primary importance is compatibility – if the software or portable player you want to use does not support your format of choice, it doesn't matter how good it sounds. Cost is a non-issue, as all our tests were performed with free software, and there are plenty of others out there. Encoding time is similarly unimportant. While there is quite a lot of variance in the speed of encoders, they're all very fast, and the speed of your CD-ROM drive will have more to do with your overall ripping speed than the speed of the codec. Product:

MP3/LAME 3.95.1

Web site: http://lame.sourceforge.net Pros:

Supported by virtually all software and hardware

Cons:

Horrible 64k sound quality, mediocre quality at other bitrates, getting the best quality requires seeking out a top-notch encoder

Summary: Near-universal support is MP3's strong suit, but the quality leaves something to be desired. Score:

Product:

12 of 14

Windows Media Audio 9/Windows Media Encoder

10/3/2006 16:05

Audio Codec Quality Shootout

http://www.extremetech.com/print_article2/0,1217,a=123493,00.asp

Web site:

www.microsoft.com/windows/windowsmedia

Pros:

Broad device support, best 128k and VBR sound quality

Cons:

Doesn't play on an iPod, the most popular portable player

Summary:

As long as you don't own an iPod, this is a fantastic choice

Score:

Product:

Ogg Vorbis 1.0.1/Vorbis-tools

Web site: www.vorbis.com Pros:

Totally free, no royalties or licensing costs, best sound at 64k

Cons:

Only supported by a few portable players, and not the most popular brands/models

Summary: Ogg Vorbis' sound quality is good and the price is certainly right. If the demand is there hopefully more devices will add support Score:

Product:

PEG-4 AAC/iTunes

Web site: www.apple.com/itunes Pros:

Supported in one of the most popular music apps (iTunes) and the most popular brand of player (iPod), good sound quality

Cons:

No VBR support in iTunes, very little hardware support outside iPod

Summary: The sound quality is there, but iTunes needs to add VBR support and more players need to support the format. Score:

Both AAC and Ogg Vorbis made a very strong showing here, and we have no problem recommending either one based purely on sound quality. Unfortunately, they both have limited adoption in hardware devices, which limits their usefulness. AAC has an advantage in that one of the very few devices that support the format is the iPod, which enjoys a huge market share. Ogg Vorbis' main advantage is sound quality, especially at the low 64k bitrate. MP3 is everywhere, and you can play back your MP3 files on virtually any portable music player or with any audio application, regardless of platform. It's quite simply ubiquitous. We searched far and wide to find the best-sounding encoder around and believe that the latest build of the LAME code is it, yet even then the sound quality didn't quite stand up to the competition, and 64k files sound like they're being played back through a tin can and string. If you want to make sure your digital music

13 of 14

10/3/2006 16:05

Audio Codec Quality Shootout

http://www.extremetech.com/print_article2/0,1217,a=123493,00.asp

can be played anywhere, MP3 is the way to go -- otherwise it should be avoided. WMA9 seems to hit the sweet spot between quality and compatibility. There are a great many popular applications that support the standard for both encoding and decoding, and all the popular portable players support the format. It often scored highest in our listening tests, and when it didn't, it was close. The main drawback is compatibility with the iPod – Apple's player is the most popular on the market, yet it does not support WMA. This is the codec we would choose if iPod compatibility wasn't an issue. If you're not planning on buying an iPod, then this consideration won't trouble you. Where does the future of audio codecs lie? It's hard to say, exactly. With ever-larger storage growing cheaper and more compact with every passing year, it's tempting to think that mathematically lossless compression will become de rigueur. Certainly we expect to see it supported by more devices. Still, it probably won't become standard – or dedicated music players might grow until the low end represents tens of gigabytes, making lossy compression unimportant, but soon cell phones and PDAs will become music repositories as well, and will have far less storage. As DVD-Audio and SACD formats continue to gain traction, and more recorded becomes available in high-resolution (24bit / 96KHz) and multi-channel (5.1 surround) formats, compressed audio will have to offer these features as well. WMA already does, while others, like AAC and Ogg Vorbis, are partway there with 5.1 capabilities – it only needs to be exposed in today's popular applications. These 24/96 and 5.1 audio files would be way too large, even for hundred-gigabyte devices, when losslessly compressed or uncompressed. Lossy compression is here to stay, and steadily evolving. Copyright (c) 2006 Ziff Davis Media Inc. All Rights Reserved.

14 of 14

10/3/2006 16:05