----- Original Message -----
From: "vdbj" <[EMAIL PROTECTED]>
To: "Caster" <[EMAIL PROTECTED]>
Sent: Monday, May 15, 2000 1:41 PM
Subject: Re: [MP3 ENCODER] VBR Bug Report
> Hello Caster,
>
> Monday, May 15, 2000, 12:23:34 AM, you wrote:
>
> C> For encoding I used lame 3.82 from http://www.chat.ru/~dkutsanov/. For
> C> decoding, Exact Audio Copy with Radium fhg codec 1.2.
>
> be aware that this 1.2 plugin has very poor performance on VBR. (last
> half second is missing etc), to be sure just decode back to PCM or wav
> and then do a freq analysis.
I tried to decode with winamp and even with lame, same results
>
> C> First I encoded the file with lame -V 1 -b 128 -h -m j
> C> After decoding and opening in CoolEdit there are clearle visible volume
ups
> C> and downs in waveform view. Freqency analysis that the frequency curve
is
> C> not almost almost straight like in original file but there are also ups
and
> C> downs above 16 khz. When listening, there are very significant chirps
(as
> C> vdbj decribes them in his mail from 13.5.).
>
> C> The most interesting thing is that vdbj's sweep produced with Sound
Forge
> C> (described on http://www.r3mix.net/) was clean! And it has the same
> C> frequencies as mine sweep, even higher.
>
> I don't know about this A440 setting, but when you do a complete
> 40Hz-22kHz sweep in CEP (instead of the 4.5kHz-22kHz SF sweep I used),
> I got some visual minor distortion in the +17kHz range, but nothing
> audible so not important.
I used 20hz-20khz sweep with log sweep on (in final settings) nad the
distortion were not minor and were audible
>
> (some part out of another mail:
>
> I used: generate tones:(B/M/Mf): initial (0/0/10) final (22050/0/0), dB LR
> -1.4, 10 seconds long
> (I wanted (again mistake: (0/0/0)->(22050/0/0)), but same outcome)
>
> So not much different conclusion:
>
> 256: perfect
> vbr: audible perfect, gfx some distortion
> xing: major cut-off at 16kHz)
>
> Seems to me there are 1001 ways to do this kind of sweeps, and the
> only way I used them was to easily show flaws in FhG vbr and Xing.
But on your page, you use sweep that VBR mode handles perfectly, I was able
to reproduce it too, I don't know why but my sweep is encoded badly, and
looks like music too...
> When I listen to what lame does in vbr mode, the only flaw I can
> detect is that with some cymbal sounds not enough bits are assigned
> (imho). My idea is just: for the moment 192kbit/s sounds poorer to me
> on this area than 128 because @128 the lowpass filter gets out
> potential sources of artifacts while @192 the full range is not
> encoded too well I think.
as far as I know VBR mode in lame uses no lowpass filtering, and the -k
switch has no effect
> C> Testing with some music source the analysis was also different above 16
khz
> C> (as vdbj reports in his mail).
>
> You cannot expect a perfect exact graph from original source and
> encoded wav. I know 256 comes close, but this doesn't mean that a VBR
> mode having more visual abberation from the original has poorer sound
> quality.
But I expect that CBR256 mp3 will produce same results as VBR forced to use
only 256 frames... and it isn't, CBR is perfect and VBR does errors.
Well, look at this page - http://www.mageo.cz/home/CASTER/lame.html and see
what i found, I made it now, very quick so there are not all the tests i've
done...
And last thing, my friend now managed to reproduce my test (generated his
own file with CEP, encoded as I did and it was same as by me)
Bye, Caster
> --
> Best regards,
> vdbj mailto:[EMAIL PROTECTED]
>
>
> --
> MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
>
>
>
--
MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
