192khz 48khz recording

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.

kambo

Well-known member
Joined
Apr 24, 2009
Messages
1,975
Location
CA
recorded exact same audio source at 24/48khz and 24/192 khz (sample library sound from kontakt)
time stretched both 100% ( yeap, can hear the glitch on both files)
exported both files at 48khz (my clients doesnt like anything higher than 48khz)
lined up in cubase-phase reverse on one.... hmmm its a null  :eek:



 
If i read correctly, we're talking about a recording of a bandlimited file, not a direct capture?
 
that's what i understood , so the new recording can't get better than the original sampled material.  Most sampled libraries are not done 24/192
 
i will try do with live recording.
(will mike up a speaker and play some MIDI notes/song from a sequencer)

but, even with sampled libraries; when  re_recorded at higher sample rate = more sample point / data for
time stretch algorithm? that doesnt make it sound better but more intense data to work with, which is better for the end result ?

 
ubxf said:
[...] the new recording can't get better than the original sampled material [...]
Exactly.

Analogy 1: if you take a web picture of 72 dpi and make it two times as big, it's still low resolution. Only difference is that one green pixel has been blown up to 4 exactly the same green pixels. When looking at the blown up picture from a greater distance it will look exactly the same as the smaller one from closer by. The same happens when upsampling material recorded at lower sample rates.

Analogy 2: take an 128 kbit/s MP3 and turn it into a 24bit 192Khz file. If you can hear a difference between the two files, you sure have magical ears.

However, if you intend on really heavy ITB processing, upsampling is not necessarily a bad idea. Yet the result might easily get thwarted by bad downsampling software again when converting back to 48 or even 44. That being said, the effect of recording acoustic stuff (direct capture) at higher bitrates is more easily discernible.
 
i made some further recordings today for extreme time stretching... 
i thought higher sample rate would be a significant improvement, but
depends on source sometimes 16/44.1 sounded the best...
24/48 is just fine....
 
i do a lot of pitch shifting (several octaves down)of sound effects recorded in the field. when you work with higher sample rates vs 44.1 you quickly notice the difference. Also when you are mixing several tracks even though the individual element might be 16 bit 44.1 the mix session itself sounds better to me at 24 bit
 
kambo said:
but, even with sampled libraries; when  re_recorded at higher sample rate = more sample point / data for
time stretch algorithm? that doesnt make it sound better but more intense data to work with, which is better for the end result ?

I have a hard time explaining why I don't think that's the case, but I feel intuitively that there is no benefit.

In a nutshell; the samples are numerical values that represent a signal that doesn't exist until it's reconstructed. So I'm betting that by upsampling you're actually not adding any new information at all, just adding more sample points. You can think of it as having a two-dimensional 'canvas', a piece of paper. You draw a perfect line between two data points, and then double the data points to four. But since the line is perfect to begin with, what do you really gain?

In other words; any processing you would do to the "line" using two data points would give you the same results as using four, because the line itself is the same. No new information.

Since increased "resolution" really corresponds to an increased range of frequencies I would then expect to see a difference only if you're gaining new information when going to a higher sample rate - i.e. higher frequencies. But you're not generating higher frequency content by upsampling your signal, so.....
 
Reverse test:
(1) Take a live capture (eg. a cymbal) and record it at 44Khz or 48Khz.
(2) Record the same sound source at 192Khz.

Now pitch down (sorry. I was thinking tape here) timestretch both file by the equivalent of an octave,  then by two octaves, three octaves, four etc
Do you hear a difference?

 
after, certain amount of timestretch, changes in transients becoming more audible!
also slight change in loudness, and focus.
cant say which resolution is better tho.  i dont feel like changing to higher sample rate.
as i sad, 24/48 is just fine.
 
Script said:
However, if you intend on really heavy ITB processing, upsampling is not necessarily a bad idea.

Non-linear processing plug-ins, such as compression and distortion effects (OK, compression is distortion) will upsample and do the processing at a higher sampling rate because the processing can result in harmonics that would otherwise fold back into the audible range due to Nyquist. The downsampling back to the original sample rate is done such that the harmonics are filtered out.

-a
 
Yes, also called aliasing. Visible in spectral analysis.

Can't remember where I read this but some famous electronics dude suggested opamps best have a non-filtered working frequency range of up to 500Khz(!) to prevent non-linear distortion from becoming somewhat/somehow audible. That's quite a number. Anything less than that -- I'd then naturally assume -- is called *mojo* ;) And anything higher than that sure must be *stardust* ;)
 
Script said:
Yes, also called aliasing. Visible in spectral analysis.

Can't remember where I read this but some famous electronics dude suggested opamps best have a non-filtered working frequency range of up to 500Khz(!) to prevent non-linear distortion from becoming somewhat/somehow audible. That's quite a number. Anything less than that -- I'd then naturally assume -- is called *mojo* ;) And anything higher than that sure must be *stardust* ;)
This is due to trying to avoid inter-modulation distortion (which may end up in the audible band), and to keep phase shifts in the audible band to an absolute minimum.
 
Trying to evaluate sound quality by altering the time base is an unnatural act. One of my early technician jobs was for a company involved in pitch shifting tape recorded speech for higher speed playback aka "time compression" (used with talking books for the blind etc). The machines we made could also slow down playback "time expansion" (with restored pitch) for improved intelligibility, for language study or understanding garbled recordings.

These generally take sound samples 20-100mSec long, stretch or squeeze them to shift the pitch the desired amount then re-splice them together. Pitch shifting down means you have more stretched data than time to display it in so must discard or overlap data (discarding sounds better). Then another tricky part is how the samples get spliced together for least audible perturbations (zero crossing, and slope matching between samples helps). For time expansion you end up with gaps between samples, so time corrected samples can be repeated to fill the gaps.

There is so much stuff going on around pitch shifting that I doubt this will tell you much about sample quality. A higher bandwidth capture of signal with higher frequency content (like cymbals) will have some previously inaudible content shifted down to be audible, but so what? This is not a natural situation. It is perhaps justified to use a higher sampling rate if planning to pitch shift down but this is all subjective.

JR
 
Script said:
Yes, also called aliasing. Visible in spectral analysis.

Can't remember where I read this but some famous electronics dude suggested opamps best have a non-filtered working frequency range of up to 500Khz(!) to prevent non-linear distortion from becoming somewhat/somehow audible. That's quite a number. Anything less than that -- I'd then naturally assume -- is called *mojo* ;) And anything higher than that sure must be *stardust* ;)
I don't doubt that somebody said that, but to make such a broad sweeping claim about all op amps is not supported by reality.  There are many modern op amps with gain bandwidth in the GigaHertz range so 500 kHz is easy lifting.  A quick glance at TI shows one with 7GHz GBW (18,000 V/uSec).

That said back in the 70's when I was designing some BBD delay lines I had to deal with slow op amps of the day's inability to handle clock glitches in the output sample stream coming from the BBD. My solution back then was to place a real LP pole at the BBD output to scrub off enough of the HF content that the op amps could finish filtering the rest out.

Audio is nominally 20-20kHz. Good practice is to provide at least 20 dB of loop gain margin @ 20kHz (open loop gain above closed loop gain). For higher closed loop gains common op amps can run out of loop gain.

JR
 
JohnRoberts said:
Audio is nominally 20-20kHz. Good practice is to provide at least 20 dB of loop gain margin @ 20kHz (open loop gain above closed loop gain). For higher closed loop gains common op amps can run out of loop gain.

(For the newbies: ) And that is what matters. For unity-gain buffers a super-speed op-amp isn't generally necessary, but when you're building a gain stage, you need to make sure that the loop gain is sufficient for the desired closed-loop gain. The old, slow op-amps didn't have high gain-bandwidth so if you tried to make a high closed-loop gain stage, the signal bandwidth would shrink.
 
Matt Nolan said:
This is due to trying to avoid inter-modulation distortion (which may end up in the audible band), and to keep phase shifts in the audible band to an absolute minimum.

I think people should actually loosen the boundaries of audible sound a bit. When doing a simple tone sweep I can't hear above 17K  as a guy in his 40's. My kid says he might be able to "feel" the tone a bit above 18K if its loud enough.  But that's still 2000hz away from 20K. 

Does it really matter if there is inter-modulation distortion around 19 or 20K anyhow? Maybe thats the "air" people hear sometimes. Maybe crappy converters with bad filters will come back like vinyl did!

P.S. Anyone going down the sample rate rabbit hole should read Dan Lavry's papers.

http://www.lavryengineering.com/pdfs/lavry-white-paper-the_optimal_sample_rate_for_quality_audio.pdf
 
Back
Top