Subject: Re: Frequency response was Re: [linux-audio-user] Audiophile CD's
From: Jason (hormonex_AT_yankthechain.com)
Date: Mon Jan 28 2002 - 07:59:10 EET
On Mon, 28 Jan 28 you Ross wrote:
> > > Well, if a sound outside of my hearing range affects a sound inside my
> > > hearing range, I don't need to study the sound that was too high - if I
> > > could hear artifacts of the interaction, I can simply study the
> > > interaction.
> >
> > I was under the impression the resonant harmonics, artifacts etc. can be the
> > result of unheard as well as heard sonics interacting and that the playback
> > (percieved sound) of subsequent recording would suffer if that data was not
> > included with the recording or alter by adding it after the fact. similar to
> > the phono preamp mentioned earlier for the "vinyl warmth" effect.
>
> I'm not an acoustic enginner and I'm not a sampling theory expert. But this
> idea just doesn't jive with the way waveforms work. If the following analysis
> is wrong, someone _please_ correct it.
>
> Let's say I have some waveform A that I can hear, and some waveform B that is
> too high. Let's say that you are capable of identifying A (it's a simple 440Hz
> sine wave). Now, assume that sounding A and B together produces an audibly
> different sound. Since it is distinguishable from A, and B is inaudible, the
> tone of A and B must be a different waveform (call it C; it's equal to A+B).
> Since I can hear C, it must be below 22kHz. By the Nyquist Theorem I can sample
> this waveform at 44.1kHz and capture it completely.
Close, hearing is actually kind of strange, and there are different parts
of our ears that pick up different frequencies in different ways. I'd like
to be able to explain that more clearly but it's a lot of Voodoo that I
don't have a very good understanding of.
That having been said, the Nyquist theorem not only applies to
fundamentals, like A, but also to overtones like B. So if A is a
fundamental, and B is an overtone creating the composite waveform C, and
both a and b are are sub 22.05khz, then theoretically sampling at 44.1
will accurately
capture the frequency of both a and b, and therefore the correct frequency
and C. Timbre however, is a function of waveshape not frequency; the
closer a signal
gets to the Nyquist cutoff, the more generic it's shape becomes. Think
about it, if you're only sampling a signal twice per cycle, then all
you're really describing digitally is a triangle wave with the same
frequency. The lower teh frequency, the higher the resolution of the
actual waveforms shape.
Therefore, higher sampling rates are desirable because they do improve the
accuracy of the repesentation of the shape of a waveform. However, past a
certain point, and IMHO it's somewher around 40k, you reach the point of
diminishing returns. Due to the logarithmic nature of hearing, an octave
difference in pitch is a doubling in frequency, so while those
higherfrequencies do effect timbre and are therefore valuable, the amount
of actual usable information develops an inverse relationship with the
amount of data you have to accumulate in order to represnt it. I
personally think 20 bit/ 96kHz sounds great, and I don't see much point in
going much further than that.
Rupert Neve would disagree with me, and so I wouldn't hold it against
anyone else either.
[snip]
> [snip]
> > To capture realistically all the sound, sibilance, harmonics
> > of an instrument that would never be heard in reality under real
> > circumstances is the quest we're on?
> >
> > Or just to have something sound the way we want it to with the best quality
> > we can muster with what we have to work with at the time.
>
> 100% agreed!
>
yep.
-- YankTheChain.com - You can pretend we're not here. That's what I do.,
This archive was generated by hypermail 2b28 : Mon Jan 28 2002 - 07:50:33 EET