[vorbis] Vorbis vs Speex

noprivacy at earthlink.net noprivacy at earthlink.net
Mon May 26 13:58:44 PDT 2003



> nen> of thing....  I wouldn't bother.  Speex isn't that good.
> You seem to forget that Speex is exclusively for voice. So if you tested
> on anything _but_ speech it's inferior to any of the above mentined
> codecs (I'm not sure about someone singing a capella, but I'd guess it
> would be a bit better).
> It's not a low bandwidth general purpose codec. So don't try to use it
> for an unintended target and then criticize it.

Alright, I accepted your chastizement and ran some more tests.

For the source I used:

Bill Cosby - 9th street bridge (vocal stand up comedy 160k mp3)
Bill Cospy - Noah (vocal stand up comedy 128k mp3)
Slinky commercial (mostly woman singing 128k mp3)

I converted them to WAV.  The originating bit rate was vastly higher than
the final rate, so the source originally being an mp3 was not an issue.

(For the conversion to mp3 & Speex, I let it downmix the stereo to mono.
Considering the sources, I doubt there was much actual stereo component in
there anyway.)

I used LameDrop at 8k cbr.  It's lowest setting. And a quality of '0' (its
highest.) I also tried a few at 16k cbr but only a couple.  I didn't try to
tweak any further settings.  (Such as limiting the bandwidth, etc.)

For SpeexDrop, the choices were a bit harder to figure out.  Usually a
'quality' setting of 0 or 1 gave a file size somewhat near the mp3's were.

I also tried doing the ABR and VBR settings but I didn't seem to get any
consistent results.

Okay... the results...

With quite a bit of playing with the Speex settings, I *was* able to get
results that sounded slightly better than mp3's at 8k cbr.

However, the results were *not* significantly better.  You could tell that
they both sounded vastly different from the original, but they were just
different, not necessarily significantly better.

Maybe the problem is the SpeexDrop settings.  I wasn't entirely sure how to
adjust them to get the best quality or even settings comparable to what the
LameDrop mp3 settings were.  I also didn't see any settings about doing a
constant bit rate, so the tests were mostly ABR & VBR against mp3 at CBR.
Maybe with better (or clearer!) settings in SpeexDrop I might have been able
to do better.

I didn't get around to trying Ogg at ultra low settings.  Nor any other
audio format.  To be honest, I just don't care that much to put that much
effort into it.

So maybe my previous comment was not entirely accurate, but I would have to
say that it was at least 50% right.

Speex may (or may not) be technically superior to other formats.  I don't
know its theory.  (I don't know mp3's math, either.  I just listen to them,
not develop them.)  All I can say is that in the few tests I've done, the
end result did not sound significantly better than the antique mp3.  Just
different.

<p><p><p>--- >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 'vorbis-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.



More information about the Vorbis mailing list