Hello people.

I know, many blind tests have been done, and many have determined that quality setting 6 is enough to achieve transparency. But I've found a piece of music where not even quality level 9 is sufficient.

Here is a test sample download (two clicks and a captcha code): http://rapidshare.com/files/58465886/Retrosic-cut.wav.html
A wave sample containing the last 20 seconds of Deathdealer by The Retrosic. This is ripped from CD.

The encoder versions I used: libVorbis 1.2.0 as well as aoTuVb5.

Encode this sample with Vorbis at any setting you like and compare. Vorbis creates some really distinct, almost rhythmic disturbances in this "atmospheric" synth sound. The strongest they are at seconds 12-14. Up to quality level 6 (216 kbps) the distortions are lucid, but in my tests, with headphones on and full concentration, even with quality=9 (328 kbps) it was always possible to identify the Vorbis version. It got tough at quality level 10, where my success rate dropped to ~80%.

But there I did arrive at a point where testing got surplus; first because the bitrate is so high that a lossy codec is inappropriate, and second because I got transparency for this little slice of music with MP3 at 202 kbps VBR (Lame 3.97 preset extreme, the only setting I tested).

What might be so special about these sounds? For the rest of the songs that I compared, I found quality 7 to be just perfect, and I was surprised when I came to the end of this song and noticed obvious distortions.

I imagine you've ABXed this?

Yes, first just using my audio player (on random, closing eyes) but then I got really scientific with WinABX.

Good. You didn't mention it explicitly, and sometimes we get noobs who don't know about proper listening test methodology around here.

Are you sure this is ripped from CD?

Regardless, the clip's spectrogram makes it clear it was an MP3 at some point (mastering engineers on crack?):


I ran it through aucdtect and it didn't give me a straight answer - "Could not qualify the source of this track."

This is likely a transcoding issue.

Also, could you clarify what exactly this distortion is you're hearing? Is it sort of a teeth-brushing sound? Because I don't hear any difference beyond -q 3. I might just not be looking in the right place.

Well, that may be so! I don't know about the drugs , but that it may have been MP3 somewhen. This band makes purely electronic music with use of synthies and computers, they also might use MP3 (or OGG) samples. (I actually know of a semi-professional music editing software that comes with huge libraries of Vorbis samples.) But the source where this Wave file was ripped from (and encoding was done) was Audio CD for sure.

[OT]Can you know this from the looks of this spectrogram picture?? Do other lossy encoders have specific patterns too? I'd love to learn more about that. [/OT]


What an ingenious piece of software. Found it at true-audio.com. Then I wanted to run aucdtect on the whole track, but unfortunately (in the face of shortcoming hard disc space) I had deleted all the wav files except of this one sample. And gave away the disc later soon.
But if you think a test would be enlightening, I could borrow it back or do the test at my friend's computer.


Teeth-brushing... hmmmm.... could be so. For me it's hard to describe in words. Where this spheric background noise should be absolutely clear (against the "pinging" noise), it is not. This distortion is totally obvious at very low quality (probably anyone would hear it). At -q 6 it's much softer, but still very noticeable.

Should I upload my ogg encoded files?? Maybe I'm doing something wrong, but I don't believe so. Latest version, no specific settings other than -q X

MP3's "signature" is the tell-tale lowpass at 16kHz.