You and Tony seem to be identifying a discernible difference that doesn't seem to be supported by the study. Why is this?

Good question. Hopefully I can provide a good answer...

I just got done commenting in a different thread that I can only hear a difference on certain songs. On most material, I can't distinguish a 128kbps MP3 from the original wave file. But there are a few songs where I can hear definite aliasing on very high frequency material. It's subtle, but it's there. The general effect is a certain harshness to cymbal crashes. In some passages with very complex material (multiple cymbal hits combined with vocal sibilance combined with high-frequency guitar notes, a surprisingly common occurence in Rush songs, go figure), I can hear the cymbals actually break down and begin to show obvious aliasing artifacts. It's sort of a swirl and chatter sound. If you're familiar with guitar effects, it's like having a very subtle flanger effect applied to the cymbals. The effect is similar to what you get when you listen to a casette tape that's wearing out, only more subtle.

Another very specific example I can cite is at the beginning of the song "All the way from Memphis" on Brian May's "Another World" album. The song has a section in the beginning where there is some recorded crowd cheering that fades into the song. The cheering was added into the recording as an effect (it's not a concert album), and they've already processed the cheering sound with some delay-based effects such as chorus and detune. Actually, it's the sound of whistling wind, combined with a car engine, combined with the cheering fading in as the song starts. It has a very specific sound that's nice and smooth on the original recording. But when I encode it at 128kbps-fixed with Fraunhofer, the resulting MP3 file very clearly demonstrates aliasing. The cheering, as it peaks in loudness, sounds staticy and harsh instead of smooth. There's also more "pink" to the white noise, making it sound more like a jet engine than a crowd cheer. The same thing can be heard at the end of the song, since the same cheer recording is used there. It's subtle, but audible if you know what to listen for.

The reason this happens is because of the nature of data compression. High frequency sounds and white noise translate into digital numbers that are almost completely random from a compression algorithm standpoint. Compression depends upon repeating patterns, and these sounds are the exact opposite of what the compression algorithms like. It's the same problem with video compression, except with video, high frequency is defined as high contrast. That's why you always see odd artifacts around sharp edges in compressed video. The audio artifacts are exactly the same, they're just interpreted through a different one of our senses (hearing instead of sight).

But like I said, in order to hear the artifacts, you have to know exactly what to listen for, and you have to know the song well, and you have to directly compare it to the original material.

Unfortunately, I happen to know my Rush material so well that those kinds of things stick out like a sore thumb when I do hear them. It's kind of ironic: You only notice it on your favorite songs, but you notice it precisely beacuse they're your favorite songs. So it makes it that much more annoying.

That's why I bought AudioCatalyst the other day. So I could do the extra high-quality VBR encoding on those particular albums to take care of it. The rest of my collection is fine at 128k, and I intend to leave it that way.



Tony Fabris
Empeg #144