25 Nov
2015
25 Nov
'15
1:34 p.m.
Again, you're missing several obvious points, all of which have been made
before.
Nyquist says that to **perfectly** capture all frequencies up to N kHz, you
need to sample at N*2.
Human hearing ends even in the best of us at about 20kHz. We sample at
typically 40kHz plus.
The implication: this "low resolution" example you're talking about
isn't
remotely close to what is happening with real world digital audio. Sure, if
you sample at 8kHz like early digital telephony (thus only capturing
frequencies up to 4kHz, the result after D->A differs substantially from
the original. But that's why we don't do that.
Also, your drawings ignore the basic point that Felix made. Your "original"
curve includes frequencies that are above the cutoff for the "sample rate"
your "digital" version uses.
On Mon, Nov 23, 2015 at 2:55 PM, jonetsu(a)teksavvy.com <jonetsu(a)teksavvy.com>
wrote:
On Mon, 23 Nov 2015 09:46:55 -0800 (PST)
Len Ovens <len(a)ovenwerks.net> wrote:
I watched the video. I still remain with what I said, really. As a
parallel, a picture - parallel also made in this video - if you take a
very low resolution of the Mona Lisa it will not be possible to
reconstruct the quality of the original. The information is not
there. This is quite obvious with pictures. Same with sampling.
Attached is a quick drawing to illustrate. It is coarse as there are
very few sampling points, but the principle applies. The limit is the
finesse of the source material. If there are extremely minute changes
in the wave then the sampling must be done at an according rate if one
wishes to preserve the original as much as possible. Parameters of
interpolation can be applied, but how would you interpolate a 4-bit
Mona Lisa to show the subtleties of Da Vinci's painting ?
Now, I would like to know that in reality things are not like that and
that high, true-to-orignal, quality can be made from low quality sources
so I remain open to all comments and ideas.
_______________________________________________
Linux-audio-user mailing list
Linux-audio-user(a)lists.linuxaudio.org
http://lists.linuxaudio.org/listinfo/linux-audio-user
On Mon, 23 Nov 2015, jonetsu(a)teksavvy.com wrote:
> On Mon, 23 Nov 2015 17:00:47 +0000
> Will Godfrey <willgodfrey(a)musically.me.uk> wrote:
>> Although it seems counter-intuitive that
actually is quite wrong!
Hmmmm.
Not sure about intuition. A sine wave with 16 sampled points
will end up like a linked list of edges. Saving that to file and
resampling at 128 will only add points to the straight lines. It
will not create curves as per the original. It cannot. How would
it know it was a sine wave and not a guitar tone when it had to
process basically what was a robotic tone ?
Very simplified.... Filtering. Really, watch the video a xiph.org
before you say any more. They compare (on analog equipment) input and
output wave forms at 44.1k.