On Tue, Oct 11, 2011 at 11:42 PM, Atte André Jensen
<atte@email.dk> wrote:
On 10/11/2011 11:36 PM, Fons Adriaensen wrote:
There are at least two good reasons why such a test is
not valid.
1. The difference signal doesn't tell you anything about
audible differences between two signals. It's fairly easy
to make a linear filter (no compression involved) with
a perfectly flat response and that nobody would be able
to hear. But when you take the difference between in and
out it is 3 dB higher than both.
2. Lossy compression is based on combined temporal and
spectral masking - some signals you can't hear in the
presence of others. Of course when you take away the
masking signal they become apparent...
BTW: Apparently the guy in the link has lifted the idea from this clip:
http://www.youtube.com/watch?v=wzJbjHc6bRE
where George Massenburg does the same. I don't know George Massenburg, but he supposedly is some kine of authority in the world of audio engineering. Not that it makes the test more valid...