[speex-dev] Way to measure loss of quality

[Jean-Marc Valin]

I very well understood your idea. The problem is that while it works for
apples, it doesn't work for speech because the amount of distortion
percieved is not directly related to the difference signal. The ear is
much more complex than that.

If you reall want to try the experiment. I can already predict some of
the results. For example, you'll find that by looking at the difference
signal, perceptual enhancement makes the signal worse, while in reality
it makes it sound better. Also, you'll see a "huge" difference between
wideband quality 5 and 6, while in reality the difference is smaller.

        Jean-Marc

<p>Le mer 20/08/2003 à 03:39, Thomas S a écrit :
> I think you misunderstood my quality measurement idea.
> I mean if you subtract the original and the one after,
> the LESS voice that is less over or the LESS you can
> tell when someone is speaking, the better the
> compression. This is still subjective but I think its
> easier to tell this way because its easier to tell how
> much voice is remaining than to tell how much the
> compressed voice is missing from the original.
> 
> Quick analogy, if you have 3 huge piles of apples (one
> being the original), of various colors. It is very
> difficult to tell which of the other 2 piles is
> closest to having the same proportions of each color
> BUT, if you subtract them out, it becomes very easy to
> look at it and see which is closest to the original.
> Hope that makes some sense, but I fully admit it,
> until I give it a try some day down the road, I can't
> be confident in this idea.
> 
> 
> > What do you prefer: a codec that sounds good or one
> > for which
> > subtracting the original gives a better signal? 
> 
> > > QUALITY MEASUREMENT IDEA:
> > > I find it difficult to hear 2 voice samples and
> > tell
> > > which is nearer the original, especially if the
> > > background hiss is slightly different. So what if
> > you
> > > actually subtract the post-compression sound from
> > the
> > > original and then listen to the DIFFERENCE. If you
> > > can't hear any voice except background noise and
> > some
> > > hiss from "s"'s and "d"'s and such, that means
> > most of
> > > the actual voice has been maintained. If you can
> > hear
> > > someone speaking, that means there is leftover
> > voice.
> > > If you can actually understand what that someone
> > is
> > > saying, the difference is great. Thoughts?
> > 
> 
> 
> __________________________________
> Do you Yahoo!?
> Yahoo! SiteBuilder - Free, easy-to-use web site design software
> http://sitebuilder.yahoo.com
> --- >8 ----
> List archives:  http://www.xiph.org/archives/
> Ogg project homepage: http://www.xiph.org/ogg/
> To unsubscribe from this list, send a message to 'speex-dev-request at xiph.org'
> containing only the word 'unsubscribe' in the body.  No subject is needed.
> Unsubscribe messages sent to the list will be ignored/filtered.

-- 
Jean-Marc Valin, M.Sc.A., ing. jr.
LABORIUS (http://www.gel.usherb.ca/laborius)
Université de Sherbrooke, Québec, Canada


-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 190 bytes
Desc: Ceci est une partie de message numériquement signée
Url : http://lists.xiph.org/pipermail/speex-dev/attachments/20030820/cb30639c/signature-0001.pgp