higher Khz recording

On 2005-12-16 00:05, Colin OOOD wrote:

however you have to push 32-bit a long way before it clips (around 300dB headroom or something ridiculous - UnderTow?)

On 2005-12-16 11:49, Boobytrip wrote:
What about 32-Bit sound? Well, this is actually a floating point number having a mantissa of 24-bit (including sign bit) and an 8-bit exponent. The VST format only uses the range from [-1;1] so we are not getting a full 192 dB range. FP numbers can be implemented different ways, but typically the machine resolution is around 3*10^-8 (according to Numerical Recipes).

This corresponds to a maximum dynamic range of 153 dB. Thus the advantage of floating point numbers is not the higher dynamic range, but the simplicity of using them.

On 2005-12-17 23:42, texmex wrote:

UnderTow, how did you calculate your number?... For 32-bit integer I get only 192dB headroom and for non-limited floating point 1667dB... I'm not sure if my way is correct either, though.

On 2006-04-22 20:55, generator wrote:

Firstly, higher sampling rates are ALWAYS better.

First and foremost, nevermind the frequencies heard - it allows better positioning. Higher sampling rates allow two sounds to occur closer together *temporally* and still be heard distinctly.

Tests have shown that the ears and brain can distinguish temporally (and I forget the name of this type of brain analysis) to a degree greater than the CD spec. The result would be that with a higher frequency you would hear instruments as being more distinctly positioned in the soundfield.

Here's the key: the smoother this filter, the better. Smoother means how gradually the filter is applied. If you have a very harsh filter, it is hard to avoid ringing. So you slope that filter from 40kHz up to 44kHz (for CD audio, for example). If you had room up to 96kHz you could slope it on up to 96kHz. That's the second benefit.

Third, you can also push noise up into the unaudible range and get it out of the low range. I *think* I know how this occurs, but am not sure so I won't say any more about it.

Of course, ultimately, you could make a song at 20,000MHz frequency but I'm sure there will come a point where the extra calculations don't count.

So in short, you theoretically want to use the higher frequency you can to get the best audio, up to the point where it becomes too immense from a calculation point of view.

The practical side of this would be that if you upped your audio rate to 96kHz you'd get half the instruments playing on the same speed computer as 48kHz. Half the disc space too, for sampled audio.

It's also finally debatable whether for electronic music the above is better or worse. I've seen magazines do studies where in double blind tests people PREFER 44kHz for electronic music because up at 96kHz the errors in the instruments and processing tends to be more clear.

One other note: increasing bit depth does not make the song louder, which seems to be the view of a few ppl here. It allows a greater dynamic range (loudest to quietest) to be represented. Does not make the music louder per se. The two concepts are independent.

Greater bit depth provides:
1. allows greater dynamic range which means you have a geater ability to avoid digital clipping,

One more benefit of both of the above: increasing both of the above will allow greater precision through multiple calculations. This means you might contemplate sampling at high sampling and bit rates, calculating at those bit rates AND THEN truncating or downsampling down to 44kHz for CD audio. You wouldn't get the audible benefits out the end, but during calculations you would be working with much higher precision numbers and getting less rounding error (simply put, anyway)

On 2005-12-12 23:06, index wrote:

And also humans hearing is not only about what only the ear gets...
what about the human brain,skin,stomach or all our other neurosensitive to sound parts?

Thus the advantage of floating point numbers is not the higher dynamic range, but the simplicity of using them.

Sounds can allready occure at exactly the same time. How much closer can you get than that? Wether we can distinguish these sounds is a matter of psychoaccoustics, not sampling theory.

That research suggests that the human brain can discern a difference in a sound's arrival time between the two ears of between 5 and 15 microseconds. However, 44.1Khz sampling rates give a phase (timing) accuracy of arround 2 nanoseconds. Way beyond what our brains can perceive.

Anyway, 30 µS is about 1 cm. So make sure you don't move your head when listening to the music! ;)

You are talking about the anti-aliasing filters in the analogue domain, pre sampling. Modern converters use over sampling. Usually 64 to 512 times the actual sample rate. So for 44.1Khz sampling, the anti-aliasing filters need to cut out any frequencies below 2.8 Mhz to 22.57Mhz! This is plenty of bandwidth to make a good filter in. After that the signal is converted (slowed down) to lower data rates with a decimation process. This process, due to the sync functions used, works as a brickwall low-pass filters. (The bandwidth of the sync function doesn't allow for any frequencies above the sampling rate).

So your information is partially correct but outdated.

Of course, ultimately, you could make a song at 20,000MHz frequency but I'm sure there will come a point where the extra calculations don't count.

Absolutely correct. That is at arround 40-60 Khz sampling rate. :-)

So in short, you theoretically want to use the higher frequency you can to get the best audio, up to the point where it becomes too immense from a calculation point of view.

Nope.

It's also finally debatable whether for electronic music the above is better or worse. I've seen magazines do studies where in double blind tests people PREFER 44kHz for electronic music because up at 96kHz the errors in the instruments and processing tends to be more clear.

Which magazines?

Anyway, it could be because the higher frequencies cause more distortion in amps and speakers which have limited bandwidth. This doesn't happen with music recorded with microphones simply because microphones don't pick up these frequencies to start with so they are not recorded in accoustical music.

If by this you mean that you can record signals at lower levels without being troubled by the noise floor or quantisation distortion, then yes.

It depends of the process. ALot of plugins use over sampling where relevant. No need to record at higher sampling rates for that.

On 2006-04-23 22:54, generator wrote:
I think we agree on most points.

I should reiterate what I did not make clear, which I think will clear up must of the rest. I was pointing out that a higher sampling rate will always allow more clarity, etc. and reduce error in processing. This is a fact.

HOWEVER I completely agree with you that we shouldn't all jack our computers up to 96kHz blindly, thinking there's some magic that creates better music. There certainly - and you agree with this - comes a point where the higher rates don't matter. I think where we disagree is at what point that rate occurs. No worries there, as people have disagreed before and will again

But a few responses are in order:

1. I spoke about better positioning and allow ing two sounds to occur closer together *temporally* and still be heard distinctly.

<snipped for brevity>

Hmm, I guess I was unclear in my quick typing. What I was trying to say is that in the real world, sounds can be closer together than the minimum resolution of 44kHz sampling rate.

<snipped for brevity>

I agree with your 5-15 ms numbers. However you made a mathematical error. It's not 2 nanoseconds. It's 22.7 MICROseconds, which as you can see is greater than your largest number 15.

<snipped for brevity>

My post was just designed to elucidate why, theoretically speaking, a higher sampling and bit rate is better.

Agreed absolutely. I omitted oversampling and left you with the wrong impression. Just between you and me I still have my doubts that oversampling is as accurate as sampling at a higher rate (because of the mathematical interpolation) - but that is my own personal opinion and I haven't seen a reference clearly describing the issue one way or the other.

However once you get a signal in your computer (for example) at 44kHz, it's very unlikely that you are done with it. Further processing would be better served, as I pointed out, with greater bit and sampling depths. Although those plugs can oversample and work internally at higher bit rates, my understanding with the plugin formats (and chats with developers) have led to to believe that they downsample and lower bit rate to the project format after done. So a chain of three or four plugins will oscillate from 16 bit up to 32 and back 4 times. Leaving the project at 24 bit in between will likely result a more more accurate rendition of the calculated final result.

<snipped for brevity>

I can agree with your agreement. However I have no idea (and do doubt) that it is at 40kHz sampling rate. 60kHz might be more like it, but at that point might as well move to 96 for the benefits of simply being able to half the sampling rate for legacy systems.

<snipped for brevity>

I will have to disagree. My answer was very obviously qualified by "theoretically" and I stand by it.

It's also finally debatable whether for electronic music the above is better or worse. I've seen magazines do studies where in double blind tests people PREFER 44kHz for electronic music because up at 96kHz the errors in the instruments and processing tends to be more clear.

There was a special in Keyboard magazine about 5 years ago discussing 24/96 where one of the more interesting facets was a double blind test as I described. Check it at a library near you if you like. I wouldn't put much faith in it as such (it's just a magazine) but it raised what I still think of as an interesting point, which is that higher resolution is not always better. It sometimes lets you hear things you wouldn't otherwise.

Quote:

If by this you mean that you can record signals at lower levels without being troubled by the noise floor or quantisation distortion, then yes.

You missed your calling as an editor - rewording things ...

I'm not sure though, that noise floor is avoided by having more bit depth. Noise floor is where it is, regardless of bit depth. Bit depth just gives you a greater range and better resolution at lower dB levels.

I personally would think (and I mentioned this above) that avoiding constant over/down sampling would be better.

At the least, higher bit rates are better because all the plugs I know of run at least at 24 bit depth these days, if not 32. No sense truncating back to 16 between each plug.

Damn quoting, hope I got it all correct.

Anyway thanks for the interesting chat. Certainly has me wondering how best to look into what bit rate my plugs are currently running at