[Crowbar]

Quote:

Originally Posted by gregorio /img/forum/go_quote.gif The point is that no mic pre-amp on the planet can get anywhere near 144dB dynamic range, only the very best ones can get close to 96dB.

There's no preamp involved in the example I gave! The ADC is right in the capsule, with just an anti-alias filter before it. I'm quite sure a pro-audio company wouldn't claim 130 dB if they can't meet their own specs!
If I can build a DAC I/V with noise and THD below 120 dB (as I have), I'm quite sure that professional engineers can build an analog front end to an ADC with at least as good performance.

You're incredibly short sighted by talking about limitations of typical recording and mastering processes, and missing the goal: the goal is to capture sound and then reproduce it to a point where it is not physiologically possible for a participant to distinguish it from the original performance, under any possible listening condition, and including even subconscious effects such as from the Hypersonic paper, or other effects of sound quality which are sometimes noticeable only after very prolonged experience to a system (because they are just at the conscious threshold).

 

[frankR]

Quote:

Originally Posted by Crowbar /img/forum/go_quote.gif I think you need to eliminate the possibility this is a filtering issue before you draw any conclusions.

Any comments on the frequency spectrum analysis?

I've shown cutting off ultrasonic frequencies results in distortion of audible frequencies.

 

[Crowbar]

How were they cut off though?
And the Fourier transform is done with some windowing to create those spectral plots; it's not mathematically ideal.

 

[frankR]

Quote:

Originally Posted by Crowbar /img/forum/go_quote.gif How were they cut off though? And the Fourier transform is done with some windowing to create those spectral plots; it's not mathematically ideal.

They're not cut-off, the're just not resolved. The 16/44 digitization is not recording the actual signal. It's smoothing it out, that is distortion!

I can't take accurate data of a signal that has much of its information above 20 khz with an instrument only capable of making a measurement below 20 khz.

The optic analogy is trying to resolve an object which has spatial frequencies above the resolving capability of the optics. What you get is a blurred image from the point spread function. Blurring high frequencies objects lowers the contrast of adjacent objects it can resolve. That's what happening to the audio, it's being blurred.

Have you listen to the audio samples? I'm shocked you can't hear the difference, it's becoming night and day for my ears the more I listen to them.

If you watch the graphical frequency spectrum in AVS Audio Editor in realtime you can see there is a lot of frequencies that the guitar makes that extend to 50 khz, and probably beyond. You think 20 khz is adaquate?

If I could find an app that can output the raw wave form I can import it into IDL and do some better numerical analysis.

What do you mean by Fourier windowing? Are you refering to the frequency bin? The bin sizes are very close, like I said there is some error attributed to this binning discrepency, but the second post that showed the graphical representation really gives you a sense of the difference.

 

[Febs]

Quote:

Originally Posted by frankR /img/forum/go_quote.gif If you watch the graphical frequency spectrum in AVS Audio Editor in realtime you can see there is a lot of frequencies that the guitar makes that extend to 50 khz, and probably beyond. You think 20 khz is adaquate?

A 24/96 recording cannot have frequencies that extend to 50 kHz.

Moreover, the fundamental frequency of the E on the 24th fret of the "E" string on a guitar--which is about the highest note that some guitars can play and higher than many guitars are even capable of playing--is just 1.4 kHz. The guitars on the recording you linked to are playing at least an octave lower than that at their highest. What sort of musical content do you imagine exists at 50kHz, more than five octaves above the highest fundamental frequency the guitar is capable of producing?

 

[frankR]

Quote:

Originally Posted by Febs /img/forum/go_quote.gif A 24/96 recording cannot have frequencies that extend to 50 kHz.

The Nyquist frequency is 48 khz, or nominally 50 khz.

Quote:

Originally Posted by Febs /img/forum/go_quote.gif Moreover, the fundamental frequency of the E on the 24th fret of the "E" string on a guitar--which is about the highest note that some guitars can play and higher than many guitars are even capable of playing--is just 1.4 kHz. The guitars on the recording you linked to are playing at least an octave lower than that at their highest. What sort of musical content do you imagine exists at 50kHz, more than five octaves above the highest fundamental frequency the guitar is capable of producing?

Something, according to the data.

The modes of a guitar string(s) and accoustic body are very complex.

The following article shows how many octaves with signifigant energy an open A bass guitar open string detuned to a 50 hz fundamental has. There is signifigant energy out to 14 harmonics, or 4 octives.

http://arxiv.org/ftp/physics/papers/0605/0605154.pdf

Imagine multiple strings resonating together along with the guitar body how complex the frequency spectrum is. Harmonic of harmonic of harmonics. I wasn't implying that frequencies as high as 50 khz are critical for reproducing a guitar.

22.4 khz is only 4 octives from 1.4 khz. 44.8 khz is only 1 octive from that.

 

[nick_charles]

I did a blind test on the 24/96 and 16/44.1 files and scored 13/15, but the difference was minute, real straining at gnats stuff. However when you downsample the 24/96 to 16/44.1 and compare the 24/96 to 24/96 downsampled they are not audibly different. I did two lots of unsighted tests to make sure. Thus it may be that the original 24/96 and 16/44.1 files are not fundamentally the same recording, i.e they differ in more than sample-rate and bit-depth.

 

[Febs]

Quote:

Originally Posted by frankR /img/forum/go_quote.gif 22.4 khz is only 4 octives from 1.4 khz. 44.8 khz is only 1 octive from that.

And as I indicated, the guitar on the recording you linked to is not playing anywhere close to 1.4 kHz. It's playing more than an octave below that, so 22.4 kHz is 5 octaves or more from the fundamental.

Regardless, look at Figure 3 of the article you linked me to. How much energy is shown in the fourth octave above the fundamental (880 Hz)?

 

[frankR]

Quote:

Originally Posted by Febs /img/forum/go_quote.gif And as I indicated, the guitar on the recording you linked to is not playing anywhere close to 1.4 kHz. It's playing more than an octave below that, so 22.4 kHz is 5 octaves or more from the fundamental. Regardless, look at Figure 3 of the article you linked me to. How much energy is shown in the fourth octave above the fundamental (880 Hz)?

Nevertheless, the data indicates there is frequency content there.

 

[bigshot]

Quote:

Originally Posted by frankR /img/forum/go_quote.gif I will use the following test tracks in 16-bit / 44 khz and 24-bit / 96 khz quantization that msofsi posted back in late Dec-07. Soundkeeper Recordings Format Comparison

Don't use their 16/44.1 track. Resample the 24 bit down using the proper dither.

See ya
Steve

 

[nick_charles]

Quote:

Originally Posted by frankR /img/forum/go_quote.gif http://arxiv.org/ftp/physics/papers/0605/0605154.pdf Imagine multiple strings resonating together along with the guitar body how complex the frequency spectrum is. Harmonic of harmonic of harmonics. I wasn't implying that frequencies as high as 50 khz are critical for reproducing a guitar. 22.4 khz is only 4 octives from 1.4 khz. 44.8 khz is only 1 octive from that.

The charts in that paper are pretty useless, they have no scale on the Y axis, there is no way of knowing how much energy there is on those harmonics...

 

[bigshot]

Quote:

Originally Posted by frankR /img/forum/go_quote.gif The following article shows how many octaves with signifigant energy an open A bass guitar open string detuned to a 50 hz fundamental has. There is signifigant energy out to 14 harmonics, or 4 octives.

What you aren't taking into account is that with each octave the harmonic is significantly quieter than the octave before it. By the time you get up to 4 octaves of harmonics, you aren't going to even be able to hear it. Add to that the fact that most musical instruments (with the exception of gongs and cymbals) produce fundamental tones that are at least four octaves within the upper limit of human hearing (20kHz). That means that for a harmonic to exceed the capabilities of redbook, it not only will be too quiet to hear, it's at a frequency you can't even hear. Totally unimportant.

See ya
Steve

 

[frankR]

Quote:

Originally Posted by bigshot /img/forum/go_quote.gif Don't use their 16/44.1 track. Resample the 24 bit down using the proper dither. See ya Steve

I'll try that.

 

[bigshot]

Make sure the line levels are matched when you compare the two too.

See ya
Steve

 

[frankR]

I don't need to prove that the guitar is making high frequency sounds, it's in the frequency spectrum I've already shown.

Do you even both to look at the data, or are you going to go on a tanget?

Down sampled to 16/44

Native 24/96 audio

Again, same conclusion. The down sampling distorted the audible spectrum and the details I was hearing are now gone. My ears hear it, the data shows it...