16 bit or 24 bit recording?


ramallo wrote on 4/24/2002, 10:51 AM
Hello Charlesdem;

Copy and paste from the AES web:


AES5-1998 AES recommended practice for professional digital audio - Preferred sampling frequencies for applications employing pulse-code modulation has replaced AES5-1984 and is available for purchase.

The abstract now reads:

A sampling frequency of 48 kHz is recommended for the origination, processing, and interchange of audio programs employing pulse-code modulation. Recognition is also given to the use of a 44,1-kHz sampling frequency related to certain consumer digital audio applications, the use of a 32-kHz sampling frequency for transmission-related applications, and the use of a 96-kHz sampling frequency for applications requiring a higher bandwidth or more relaxed anti-alias filtering.

ramallo wrote on 4/24/2002, 10:59 AM
Hello Chienworks,

The audible beat tone is equal to the diferential tone, you only can ear the beat, if the two interferenced tones are on the audible range.


P.D. I can have several understand errors by my bad english :( , sorry.
inspector wrote on 4/24/2002, 3:33 PM
It has been a long time since I had physics but it sounds like what is being referred to is called constructive or destructive interference. I believe for 55,000 and 55,440 to sound like 440, one would have to be the inverse of the other - destructive interference. We did this in the lab. I could also be totally off the mark.

bgc wrote on 4/25/2002, 12:45 AM
In the spirit of driving this 16/24 bit thread over the cliff, down the side of the hill and into the ocean...
If you have a 96kHz compatable sound card, try the following:
In Sound Forge create a 40000 Hz tone. Play it. Can't hear it (I can't)
Create a 40400 Hz tone. Play it. Can't hear it (I can't)
Take the two, mix them together (or make one left, one right in a stereo track and downmix to mono).
Hey, I can hear something around 400 Hz!
Interesting experiment. Quick. Fun... well not really that much fun. :)
Sorry... hey, I like 24 bit recordings!!!
nlamartina wrote on 4/25/2002, 1:36 AM
I can't believe I missed this thread! One of my favorite topics! I'm too tired to make this elegant, so here goes...

I've done a lot of research in this area, and the conclusion that I've come to is that the ultimate system setup (in interest of honestly reproducing an auditory event) must be able to respond to frequencies between 1Hz and 50kHz, +/- > 1dB, while recording a 200kHz sample rate at 24 bits (all from microphone to monitor). Now of course this isn't even slightly realistic yet, but here are the motivations behind it (some of this will be review for you):

Inside your ear, down the auditory canal, through the cochlea, and into the Organ of Corti, are thousands of tiny hair cells. They are arranged into several rows and two groups, inner and outer. As acoustic pressure waves pass over them, they vibrate in accordance to the wave characteristics. These movements, after being spectrally filtered, are converted into nerve pulses and sent to the auditory cortex in the brain for interpretation. The odd thing is that only about 20% of these nerves (3000 out of 15000) are hardwired to the cortex; they are the nerves we "hear" with (presumably, they’re the ones that respond to frequencies between 20Hz and 20kHz). It is unknown where the rest of the impulses go, but it is suspected that they create either a physical or emotional response in the listener, cued by overtone frequencies above 20 kHz (Kelley, this would explain your aversion to a 74kHz triangle wave).

Here’s some test data I drawn from secondary sources:

When a tone above 20 kHz is played to a listener, electrical brain activity occurs (measured on an EEG) despite the fact the subject cannot audibly perceive the tones. This recorded brain activity stops when the tones exceed 50 kHz. In addition to this, declassified Department of Defense tests have shown that prolonged exposure to sub-sonic noise has destructive psychological effects (maybe that’s why the morons with the booming bass cars spend some much money fixing up crappy automobiles).

Finally, other tests show that the internal audio "clock" of the human brain is around 5 microseconds. Convert that, and you come up with a 196kHz sample rate. Therefore, if you capture more data than the human ear can hear, the body can feel, and the brain can interpret, said captured data would sound indistinguishable to the source material (ignoring cross talk, positioning, frequency variance, etc).

So yeah, it's all theoretical data, but I think that's got to be the limit. Unless someone can show me why I'll need to otherwise, I'm not gonna buy a machine that samples over a 200kHz rate (provided I live long enough to see my wallet get fat enough to think of buying something that advanced).

Nick LaMartina
ramallo wrote on 4/25/2002, 2:24 AM
Hello bgc,

Have you speakers with 40400 bandwith? :-O, whow!. Try with a 21 kHz and 21,4 kHz is more real. And advice, don't forget remember the distortion, you can ear tones of 400 Hz 500 Hz, 800 Hz etc, be careful with the volume (Damage speakers).

bgc wrote on 4/25/2002, 1:37 PM
hi ramallo,
actually i used stax 404s (headphones).
ramallo wrote on 4/25/2002, 7:14 PM
Hi bgc,

Nice headphones (I love electroestatics), but I don't beleive that this headphones (And the rest of your electronic) offer 40000 kHz with low distortion for make this test, remember that you listened 400 Hz tone (400Hz, are you sure?), will be made by a distortion factors of your headphones and/or electronics. Try make the test with 21 kHz and 21,5 kHz, I put my hand on the fire that you can't ear nothing (Set the tones at low level, high level will be distortioned your equipament).

Regards and good luck.
bgc wrote on 4/26/2002, 12:57 PM
yup heard it a 21000 and 21400.
what can i say?
ramallo wrote on 4/26/2002, 4:50 PM
Hello bgc,

Sorry, I don't understand well your purpose with your reply (English isn't my native language).

You try with low tones? (21 kHz for example), you can heard the diferential tone?.

bgc wrote on 4/26/2002, 5:05 PM
Yes, I hear the beat frequency.
bgc wrote on 4/26/2002, 5:19 PM
Let's do everyone a favor and kill this thread. We're way off track now.
ramallo wrote on 4/26/2002, 8:00 PM
ramallo wrote on 4/26/2002, 9:24 PM
Hi again bgc,

For your records.

I put in internet a file of 16/48, with a 5 seconds of -10 dB 1 kHz tone, follow by another 5s of two tones of 22 kHz and 23 kHz, you can try with this test tones.

In the same test I analiced (FFT) the 5s of 22+23 kHz of this file, captured with a earthworks M30 mic and Sound Devices USBpre, reproduced by Lucid DA 9624 by an Sennheiser 580 headphones, (I set the mic inside the headphones).

The diferential tone (¿Beat tone?), the teory said: this effect is by a human hearing defect, if this is true, I can't measure the 1 kHz diferential tone (Because is a human defect).

Damn!, I can measure the 1kHz tone, and other tones, will be a product of the distortion of the headphones/electronic?, I believe that this question is true, by the high levels necesary for optain this tones.

In this web adress you will optain the test file and the captured screen of my FFT analizer. If you see the FFT results, you will see the diferent distortion generated tones

Is enougth for you?



>Let's do everyone a favor and kill this thread

I know you motives for kill this tread (Is very interesting for me and others). You don't be rigth in this disscusion, is more easy escape by the back door than offer a technical facts.
MarkWWW wrote on 4/27/2002, 10:02 AM
What you are seeing in your FFT graph is a result of non-linearities in your measuring system. In a truly linear system, when you add together signals of different frequencies you do not get these difference (and/or sum) frequences created.

Have a look at http://www.cix.co.uk/~markw/fft.jpg where you will see the results of repeating your experiment (adding a 22kHz signal to a 23kHz signal) purely within the digital realm (where things should be linear). As you will see, there is no 1kHz signal produced.

I did this by creating a pair of files using Sound Forge's "Simple Synthesis" function - each was 16-bit, 48kHz, 12 seconds in length, one containing a sine wave at 22kHz and the other a sine wave at 23kHz, each at a level of -6dBFS. In order to minimise the effect of transients I then applied a 1 second fade in at the beginning of each file and a 1 second fade out at the end of each file. I then created a new empty file and dragged and dropped each of the two sine wave files into it in Mix mode, ending up with a new file containing the results of adding the two original files together. I the selected the whole of this new file and applied Sound Forge's Spectrum Analysis function to it, resulting in the graph I have posted. Try it yourself and you will get the same results.

Whilst it is true that electronic circuitry is not completely linear (and neither is the human ear), in well designed equipment the discrepancy between the actual performance and that of an ideal (perfectly linear) system can be made very small and when used properly you should not be able to detect significant levels of difference frequency in the experiment you described. Somewhere in your chain of gear you have something which is either poorly designed or being pushed beyond its intended range of operation. (I'd guess the latter - from the look of your graph something is being pushed into distortion somewhere along the line).

MarkWWW wrote on 4/27/2002, 10:13 AM
Did you ever get round to trying this, and if so did it work?
Chienworks wrote on 4/27/2002, 11:22 AM
Actually, if i may be blunt, this whole experiment is pointless in the digital realm. You *CANNOT* produce a sine wave at frequencies that approach the even a substantial fraction of the sampling rate. At a 48KHz sample rate, any attempt to create a sine wave at above, say, 4Khz will be grossly inaccurate despite what Nyquist says. At higher frequencies all you're producing is interference patterns between the wave form and the sample rate. Try this simple experiment: create a new file with a 2KHz sample rate, then synthesize a 660Hz sine wave. Zoom in to 1:1 and see what you get. Better yet, listen to it. It's not a sine wave. Synthesize a 1KHz sine wave and you end up with a flat line. (Why? because the sampling rate measures the waveform at what just happen to be each zero crossing.) A 500Hz sine wave will end up being a triangle wave. If you want to experiment with mixing frequencies in the 20KHz range, then you'd better set the sampling rate to 384KHz. Hmmmm. Does Sound Forge support this?

That being said, the whole reason i originally brought this topic up was to point out that there are sounds that exist in the real analog world that can't be captured and reproduced digitally. Beat frequencies created in real music can be created by real frequencies too high to be reproduced digitally, and therefore won't be recreated accurately in a digital recording.
ramallo wrote on 4/28/2002, 6:11 AM
Hello Markwww,

>Somewhere in your chain of gear you have something which is either poorly designed >or being pushed beyond its intended range of operation. (I'd guess the latter - >from the look of your graph something is being pushed into distortion somewhere >along the line).

The objetive of the experiment, was show the diferential tone or beat tone don't exits at frecuencies above the human earing limit (Almost for the humans).

In my test all test was made in the analog domain, because the diferential tone only is a defect of human hearing, if you analize my test file (Made with Sound Forge), the tones are clear and without any significative distortion.

I play this files by a DAC and speakers (And mesure with a microphone), and up the sound level with the objetive of generate high levels of distortion, why?, for show that the measured diferential tone only was a distortion factor of electronics/speakers(If the DT is a human defect I can't measure, in digital or analog). I made this test with high-end equipament (Very good DAC,speakers and microphone), but with the objetive of findig the distortion tone.

My conclusion: the diferential tone don't exists above the human earing limit, for humans, of course.

Chienworks wrote on 4/28/2002, 7:00 AM
Ramallo: once again, you're changing the subject. Beat frequencies exist in the real physical world and are not dependant on human hearing. None of what i've been posting has anything to do with differential tones. Beat frequencies can be anywhere in the frequency spectrum and are not limited to the human hearing range. They are entirely due to sound waves mixing with each other and cancelling/augmenting each other.
ramallo wrote on 4/28/2002, 12:53 PM
Hello Chienworks,

> Beat frequencies exist in the real physical world and are not dependant on human >hearing.

The diferential tone is a subproduct of beat tone (to human refer). The beat tone exists, but the efect over human hearing is the diferential tone.

If you can't heard the diferential tone in frecuencies over the human hearing limit, the beat tone in ultrasonics isn't significative to human hearing, ok.

I don't change any subject, but I use the name of diferential tone for refer the effect of the beat tone over the humans (Almost in Europe the name for the subjetive third tone is diferential tone).

My conclusion is the same: the effect of the beat tone for humans is only concernig to hearing bandwith (Top limit 20 kHz), and I prove it (Almost in the bgc test, his listened tone is product of the distortion).


P.D. If you don't agree, send me a test for prove it.
PipelineAudio wrote on 6/9/2002, 3:21 PM
damned if I can get it to work

just making a hiss although it appears to read it right
bgc wrote on 6/9/2002, 6:57 PM
Hi all -
The following is a link to a paper that I saw presented at AES a few years ago and was my introduction to inner ear distortions and the creation of fcdt = 2f1 - f2.


Note: these tones cannot be "seen" using fft analysis on audio signals you generate. They are bi-products of the physical nature of your ear and must be measured within the ear. They paper presents some results in this area and discusses the need for audio coders to address frequencies, which although inaudible, may create audible components.