[vorbis] YA-2496

[Michel Donais]

> -----Original Message-----
> From: owner-vorbis at xiph.org [mailto:owner-vorbis at xiph.org] On Behalf
Of
> gtgbr at gmx.net
> Sent: 30 décembre, 2002 06:14
> To: vorbis at xiph.org
> Subject: Re: [vorbis] YA-2496
> 
> It depends on the encoding software that you choose to get your .ogg
> files from ... and it's getting trickier here. AFAIK even oggenc
> supports only 8 or 16bit integer or 32bit IEEE float input files. Mike

I wouldn't care converting to 32bit IEEE floats.... If it's all it
takes. That part is properly covered even if (like you pointed out) it
takes some time and HD space. HD is not a (too big) problem and time,
well, if that's what it takes...

<p>> About the sample rate ... well, getting a tuned mode for >48kHz could
> take really long, since that's way off normal use. You say that you
> accept quality degration; why not "degrade" quality to 48kHz and
> compress the whole thing with something lossless? If you want to make

That would mean cut off half the quality and get a lot more quantization
problems at higher frequencies, exactly what I'm trying to avoid. Some
of my work would directly be at 88.2KHz to accommodate CD halving of the
quality. But most of the sources will still be 96KHz to have the best
possible quality before the application of any filters and before
mixdown.

<p>> ... resampling would be pretty exact. E.g. if you play one of those
> sources at a different pitch, previously inaudible differences made by
> the psychoacoustic model might suddenly become audible, even at high
> bitrates. A lossy Codec like Vorbis would discard/reduce most of the
> information above 24kHz anyways, because inaudible audio information
is
> considered "not important".

That's basically the part I am afraid of. I do not care much on 24KHz
cutoff, as with the original I will probably get only a dozen "true"
frequencies on that upper range and everything else will get biased
toward these values due to quantization... and the lower 24KHz will be
much richer than on a CD/DAT source with much less quantization
problems... but I am afraid of the psychoacoustic model and what it
might do to the soundtrack, even at highest bitrate.

And then, I probably won't play the samples 4x slower than they
should... I do not do downtempo (yet).

If I continue forward with my 192KHz example, it would then be a real
loss of quality, and the model should be biased toward 48KHz or even
more (basically IMHO it should be variable with the resulting bitrate -
a 192KHz source @ 64Kbps should still be cutoff'ed @ 20 or so KHz), even
if the human ear can't perceive these frequencies at a correct rate.

<p><p>> A nice and free lossless Codec is FLAC (flac.sf.net).

This codec was also mentioned in the original 1-year-old discussion
thread. Even if lossless is a good idea for perfect archival, the
biggest problem here is the potential range of compressibility of the
source. If I have a source which consists of a few gated kick-drums to
the left or to the right, this is god-sent. But if I have a big heavy
guitar with lots of fuzz distortion and everything, and my sample is
pretty much normalized, I am afraid I won't have a lot of success with
FLAC. Or the overheads for the cymbals during a cymbal solo... Worse
comes to worst, I will fall back on this format anyways, as these are
special cases anyways.

<p><p>Have a nice day
Mike

<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.