|WMA 9 versus MP3|
In september 2002 Microsoft released the 9th release of its Windows Media technology, featuring Windows Media Audio (WMA) 9. In order to demonstrate how good its WMA is, Microsoft provided an online comparison page allowing to hear both WMA and MP3.
This comparison is featuring two test samples encoded at several bitrates. The comparison seems to be oriented toward low bitrates, as the featured bitrates are 128, 64, 32 and 20 kbps. This comparison does not include any comment, allowing the reader to judge by himself/herself, which is a good thing.
But there is a problem with this comparison provided by Microsoft: it is unfair. We could argue that the test samples could be carefully choosen in order to provide good quality with WMA and bad quality with MP3. But as there is no clue about this, we are ignoring this possibility, even if it is likely to be the case.
The problem comes from the encoding itself. The MP3 encoded files are encoded by the Lame MP3 encoder, but the version used is 3.83. This version was released in may 2000, so this comparison is featuring the just released WMA 9 versus a Lame version that is nearly two years and a half old. This is something totally unfair.
In order to be have a more accurate opinion of the respective performances of WMA 9 versus MP3, I am providing the same comparison, but featuring additional encodings. The first new encoding set is simply encodings made using an up to date MP3 encoder, Lame 3.93. The 3.93 version is used, but the results should be identical using the 3.92 Lame version, as the differences between both releases are mainly related to speed and usability. The second encoding set is encodings from Lame 3.93, but using selected additional parameters. Those parameters are selected to try maximizing quality and to reach the same filesize as the WMA samples, because the WMA samples are always bigger than the MP3 samples encoded using the same target bitrate. On the comparison page provided by Microsoft, we do not know if they used carefully choosen parameters for the WMA encodings or no.
Original sample (Baglama from The Guarneri Underground)
128 kbps samples
64 kbps samples
32 kbps samples
20 kbps samples
For low bitrates, WMA seems to be a better compression scheme than MP3. But the difference is not as huge as Microsoft would like us to believe. By listening to the sample, you will understand that recent MP3 encoders are quite better than two years ago. Even without additionnal parameters, the results are way better than the MP3 samples provided by Microsoft. By using additional parameters, the quality can even be pushed further. You can also notice that medium and high bitrates (128kbps) do not need any additional parameter than the preset value, when using a recent Lame release, in order to provide optimum quality. (note: for Lame 3.90-3.92 it was necessary to use --alt-preset instead of --preset)
At very low bitrates, it seems obvious that WMA 9 is better than Lame 3.93. But as the bitrate increases, the difference shrinks, and at 128kbps they are quite similar. As the winner in such a low bitrate test seems to obviously be WMA 9, we can wonder why Microsoft did not provided an honest comparison instead of this very unfair one.