[Vorbis] I think vorbis codec group have a new target

[Gregory Maxwell]

On Sun, Sep 5, 2010 at 3:37 AM, David.Lin <david.lin619 at gmail.com> wrote:
> I got some error,upload it to mediafire.com.
> http://www.mediafire.com/?da3lj6yhan4pad4

Of course I could produce a similar test— and I agree that it's
helpful that you posted your samples. ... but that isn't what I was
attempting to ask.

What you haven't done is demonstrated that one or the other
objectively sounds better to _you_ much less me or anyone else.

At lower bitrates or with immature or broken encoders I am willing to
believe someone's casual impression that one thing is obviously better
than another thing.  But with 80kbit/sec and a good encoder AAC and
Vorbis should both be approaching the point where many listeners have
some difficulty distinguishing the compressed copy from the original
for many sound samples.

Is it possible that you can distinguish Vorbis and AAC from the
original at this rate? Sure, but it's not certain. Is it possible that
more than distinguishing both from the original you could distinguish
them from each other? Sure, but it's also not a certainty.  ...and
that this AAC encoder sounds reliably better than Vorbis?  Again, it's
quite possible.  Each one of these cases is harder to establish than
the prior one, and we shouldn't be confident that even the first is
true.

Unless you performed a blinded test I think it's also fairly likely
that you are just imagining the difference... and I'm not going to
bother wasting my time confirming a report from someone who won't
bother to do a properly blinded test themselves.  ... Not because I
think your word is particularly untrustworthy— but because I've
discovered how untrustworthy my own casual analysis is.

So what I was looking for from you wasn't more details about how you
created the files— though thats important too— but something like "I
used this blind testing tool and gave these ratings".

For whatever it's worth the ffmpeg decode of that aac file is ~0.4dB
louder than original while the decoded vorbis file is 0.03dB quieter
than the original. I'm not sure how much difference would be required
to throw the results— but it is well known (and extensively studied)
that people tend to believe that louder files sound better.

It's normally not expected for an lossy encoder-decoder chain to make
that big a difference in the overall volume level, so normally nothing
needs to be done to control for it.  I'm not sure if it's your encoder
or my decoder which is broken, however.  If you're unable to find and
fix the source of the difference you should at least correct for it
before testing.

The my ffmpeg decode qt encoded file is also a different length than
the original and appears to be offset by about 48ms.  This all might
be the ffmpeg decoder that I'm using, so it's usually helpful to post
decoded files too.


Cheers and happy listening.