96kHz Audio

[Legion]

I was listening to a high fidelity version of an album that has a sample rate of 96kHz. I've never really bought into the idea that there's any perceptible quality difference above 44kHz due to Nyquist and hearing ranges being 20Hz to 20kHz. But I was wondering, if we can't hear something like a 40kHz tone directly, can we hear it's effects on audible frequencies?

In the attached image I added a 10kHz sine wave with a 40kHz wave. The amplitude of the 40kHz wave is attenuated by 90% relative to the 10kHz. Nothing special about 90%, I just picked it out of the air on the assumption that the frequency response of most recording equipment tends to diminish significantly above 20kHz.

The summed wave form appears distorted compared to the original sine wave. Would this distortion be audible? Or would our ears hear it the same as the original sine wave?

[cyr]

Run that through a low pass filter and you get the original 10KHz sine back.

IMHO the only advantage of 96KHz sampling is to give a good margin and not require extremely sharp filters to avoid aliasing.

[Yago]

Not sure about 96k, not something I have tried.
24 bit (as opposed to 16)does make an audible difference though!

[tszaboo]

It is because the 44100 Hz is one of the biggest lie in the history of mankind

This is happening in air, if two sine waves meet each other. Notice, that you have f1+f2 as a frequency and also, f1-f2. Now, you will not hear f1+f2 because it is too high, but I still waiting for someone to explain it to me why f1-f2 is not audible.

[Legion]

Not sure about 96k, not something I have tried.
24 bit (as opposed to 16)does make an audible difference though!

I don't think I'd want to listen to a track that actually utilized 24 bits of dynamic range. That'd be like straining to hear a whisper followed by a jet engine.

[Legion]

Run that through a low pass filter and you get the original 10KHz sine back.

IMHO the only advantage of 96KHz sampling is to give a good margin and not require extremely sharp filters to avoid aliasing.

Yeah. It has utility in an audio production context. But if you didn't put it through a filter, would the distortion be audible?

[Lukas]

Run that through a low pass filter and you get the original 10KHz sine back.

IMHO the only advantage of 96KHz sampling is to give a good margin and not require extremely sharp filters to avoid aliasing.

Yeah. It has utility in an audio production context. But if you didn't put it through a filter, would the distortion be audible?

No, because your ears have a buitlin lowpass filter at around 20kHz.

It is because the 44100 Hz is one of the biggest lie in the history of mankind

This is happening in air, if two sine waves meet each other. Notice, that you have f1+f2 as a frequency and also, f1-f2. Now, you will not hear f1+f2 because it is too high, but I still waiting for someone to explain it to me why f1-f2 is not audible.

That's mathematically correct, but if you take the fourier series of your equation, you're only going to see f1 and f2. There isn't any f1-f2 or so (effectively the left side)

[Rudane]

Not all of a sound experience is audible. In fact down around 20 Hertz and below you feel those tones instead of hear them. So, since sound is so subjective, I assumed this is where the oversampling people jumped in to increase sampling rates. I've tested my own hearing and I get nothing above 16 kHz, so technically I could probably just sample at 32 kHz for my own collection.

[DenzilPenberthy]

This is happening in air, if two sine waves meet each other. Notice, that you have f1+f2 as a frequency and also, f1-f2. Now, you will not hear f1+f2 because it is too high, but I still waiting for someone to explain it to me why f1-f2 is not audible.

notice that on the right hand side of the expression the two cosines are multiplied not added. When the two cosines are multiplied together you end up back with just f1 and f2

You get sum and difference frequencies created when you perform non linear functions such as multiplication. e.g. if you modulate the amplitude of f1 with f2 then you will also produse f1+f2 and f1-f2.

Two sine waves adding in air is linear.

[Yago]

Not sure about 96k, not something I have tried.
24 bit (as opposed to 16)does make an audible difference though!

I don't think I'd want to listen to a track that actually utilized 24 bits of dynamic range. That'd be like straining to hear a whisper followed by a jet engine.

I only used for mixing, but it was not as bad as you might think, just more room for the instruments.
(note great care to stay away from the auiodfoolery phrases!:P)

[Richard Crowley]

I always record in 24-bit if I have the option.  When recording live music, it allows for capturing peaks without clipping, while keeping the average (and even the low-level sounds) out of the "mud" (noise floor).  And when recording dialog (even scripted dialog) for video production, having that headroom is very beneficial for coping with unexpected passion from the actors (especially amateurs).

OTOH there is ongoing debate even in the classical music field (which is most of my recording) about whether to compress or artificially manipulate the dynamic range for release mixes.  Almost nobody sits in their nice quiet, acoustically-favorable room and listens to music through good speakers anymore.  Probably even for classical, the majority of people are now listening while commuting, exercising, etc. on little ear-buds of questionable quality.  Something is completely out of control when rappers (i.e. "Dr. Dre") are setting audio standards.  No thanks, Dr.D.

Of course, if you look at the loudness envelope of most "pop" music (especially hard-rock), it practically looks like noisy DC. It has the very life compressed out of it. It looks like you ran a steam-roller over the waveform several dozen times.

The sample rate for CDs was set at 44.1K by Phillips and Sony, but modern release media (i.e. MP3) make that sound "pristine" by comparison.  Film and video production is traditionally done at 48K, and for special effects and things we know are going to be subject to post-production manipulation, 96K sampling rate is often recommended.  Of course it is down-sampled to 48K for the final mix.

I submit that SACD never achieved any traction because almost nobody can actually hear the difference between 44.1K and 96K sample rate, so the cost/benefit seems significantly negative.

[tszaboo]

This is happening in air, if two sine waves meet each other. Notice, that you have f1+f2 as a frequency and also, f1-f2. Now, you will not hear f1+f2 because it is too high, but I still waiting for someone to explain it to me why f1-f2 is not audible.

notice that on the right hand side of the expression the two cosines are multiplied not added. When the two cosines are multiplied together you end up back with just f1 and f2

You get sum and difference frequencies created when you perform non linear functions such as multiplication. e.g. if you modulate the amplitude of f1 with f2 then you will also produse f1+f2 and f1-f2.

Two sine waves adding in air is linear.

On the other hand, perception of the sound works differently. It is called beat sounds (actually wikipedia article where the expression comes from). If you havent experienced it, do it, take two similar insturments, you will hear the low frequency, f1-f2. Or left speaker, right speaker. Now, if this would happen with ultrasound, it will still create a beat frequency IRL, but sadly, it is filtered out.

[Legion]

No, because your ears have a buitlin lowpass filter at around 20kHz.

Is the filter a result of the wobbly diaphragm in your ears not responding above 20kHz or because the auditory part of the nervous system has some kind of intrinsic band pass filter?

[Yago]

I always record in 24-bit if I have the option.  When recording live music, it allows for capturing peaks without clipping, while keeping the average (and even the low-level sounds) out of the "mud" (noise floor).  And when recording dialog (even scripted dialog) for video production, having that headroom is very beneficial for coping with unexpected passion from the actors (especially amateurs).

OTOH there is ongoing debate even in the classical music field (which is most of my recording) about whether to compress or artificially manipulate the dynamic range for release mixes.  Almost nobody sits in their nice quiet, acoustically-favorable room and listens to music through good speakers anymore.  Probably even for classical, the majority of people are now listening while commuting, exercising, etc. on little ear-buds of questionable quality.  Something is completely out of control when rappers (i.e. "Dr. Dre") are setting audio standards.  No thanks, Dr.D.

Of course, if you look at the loudness envelope of most "pop" music (especially hard-rock), it practically looks like noisy DC. It has the very life compressed out of it. It looks like you ran a steam-roller over the waveform several dozen times.

The sample rate for CDs was set at 44.1K by Phillips and Sony, but modern release media (i.e. MP3) make that sound "pristine" by comparison.  Film and video production is traditionally done at 48K, and for special effects and things we know are going to be subject to post-production manipulation, 96K sampling rate is often recommended.  Of course it is down-sampled to 48K for the final mix.

I submit that SACD never achieved any traction because almost nobody can actually hear the difference between 44.1K and 96K sample rate, so the cost/benefit seems significantly negative.

Hehe the loudness wars indeed!
I hate that in the 70's and 80's people aspired to having a good music system.
Certainly not all were, but many more than now.
Now the few that have are ignored for the large part.

TV/movie may have the better technical spec, but I cannot recall one recent film that had anything like a reasonable mix. Dialogue is whisper and explosions/jets window rattling... just like the film industry, all effects and little to say!

Pretty funny that a Mettalica album was brick walled to death, but fans found that the Guitar Heroes soundtrack had plenty of dynamics left in it

I hope you win out and the dynamics are left in classical recordings.
Stuff the earbud generation, they all have damaged hearing by the time they are of working age anyhow :p

[fcb]

Higher sampling rates allow for more gentle filtering in both the record and playback chains, this tends to mean more phase accuracy at the higher frequencies (44.1KHz sample rates require fairly abrupt filtering between the Nyquist and the top end of the audible band).

In practice most people can't tell the difference between 16 bit PCM at 32KHz and 24bit at 192KHz - but that doesn't mean to say that their isn't.

[Richard Crowley]

TV/movie may have the better technical spec, but I cannot recall one recent film that had anything like a reasonable mix. Dialogue is whisper and explosions/jets window rattling... just like the film industry, all effects and little to say!

Much of that problem is mixing for 5.1, but then viewing in stereo or something other than a good 5.1 system.  Some of the stereo mixes are complete failures, consisting of only front-left and front-right, and completely forgetting that most dialog is mixed to the front-center channel.  FAIL big-time.

[JuKu]

Commenting on some of the points here:
-It has been tested several times that a (very good) downgrade of a 96k/24b signal to 44.1k/16b is inaudible. The point of 96k/24b is the added headroom; you don't have to have everything exactly perfect. In other words, there is practically no room for processing, slightly misaligned gain settings, imperfections in reconstruction filters etc at 44.1k/16b, but relatively plenty at 96k/24b.
-96k/24b productions tend to be done with better and more modern equipment and be done more carefully, just because the end format is supposed to be "high resolution". The industry has a built-in quality bias for 96k/24b.
- Off subject, but somebody mentioned the physical feel: the latest AES Journal had a paper, where a music reproduction got 2-3x better grades if the seat vibrated along the music. (My interpretation: like in live concerts)

[jancumps]

For a moment I thought I was on the wrong forum (I'm also an audiokarma.org member ).

[pipe]

Not all of a sound experience is audible. In fact down around 20 Hertz and below you feel those tones instead of hear them. So, since sound is so subjective, I assumed this is where the oversampling people jumped in to increase sampling rates. I've tested my own hearing and I get nothing above 16 kHz, so technically I could probably just sample at 32 kHz for my own collection.

With some mathematical magic you can make an oversampled "1-bit" A/D or D/A converter cheaper and better than a 16 or 24-bit converter. Easier to make it linear, and most important, you get very nice benefits when sampling at a rate that's 128 times higher than your bandwidth of interest. A plain RC-filter will be enough to remove everything above the Nyquist limit, and the rest can be done digitally. Maybe this is what you mean by the "oversampling" people.

[hamster_nz]

It is because the 44100 Hz is one of the biggest lie in the history of mankind

This is happening in air, if two sine waves meet each other. Notice, that you have f1+f2 as a frequency and also, f1-f2. Now, you will not hear f1+f2 because it is too high, but I still waiting for someone to explain it to me why f1-f2 is not audible.

So what is going on is the f1 and f2 signals are going in and out of phase with each other. Where they are in phase they reinforce each other, when they are out of phase they cancel each other. The 'beat frequency' you hear when tuning a guitar is not the an actual tone, it is the amplitude modulation of the fundamental tone as they phase in and out - much like somebody rapidly tweaking the volume. This is why the right hand side is the equation for Amplitude modulation. 

So if you have a 25kHz sine wave and a 26kHz sine wave playing though two speakers you will not get a  51kHz sine wave and an audible 1kHz sine wave. On a Oscope or if you record it you will get a signal identical to a 25.5kHz signal that is amplitude modulated by a 1kHz sine wave - and it is completely inaudible.

But, if you get a 25kHz sine wave and amplitude modulate it with a 10kHz sine wave before sending it to a speaker you will get a f1+f2 of 35KHz, and a f1-f2 of 15kHz. If you correctly sample the signal at 44.1kHz after passing it through a brick wall 20kHz filter you will indeed end up with a pure 15kHz tone - everything that is audible is present. However if you filter it poorly before you sample it, the 35kHz signal will be aliased down to a 'fake' 9.1kHz signal.

The last sentence is why you are far better to sample at 96kHz when recording. As analogue filters are not perfect (slow roll-off, phase changes and so on) there will always be junk above the Nyquest limit getting to the DAC. So sample higher, then knock the inaudible junk above 20kHz off with a near-perfect brick wall digital filter before you down-sample it to 44,100. As a bonus it allows you to use a far less brutal analogue filter before the DAC.