| Author
|
downsampling from 24bits/48KHz. to 16bits/44.1KHz.
|
Boobytrip
IsraTrance Junior Member
Started Topics :
39
Posts :
988
Posted : Apr 28, 2005 17:58
|
Hi,
Normally i work at 24 bits 48KHz. when making music and at the end of the mastering i go down to 16 bits at 44.1 KHz. to be able to burn a track to cd and play it in a cd-player.
A few days ago i read a remark on this forum (i think it was by Undertow) that said that working at 48 KHz. when you are downsampling to 44. KHz. in the end has more downsides then advantages for the soundquality. I think this has something to do with the way downsampling works.
So, how does this downsampling usually work ? Are some samples simpy discarded or is there some form of interpolation happening to arrive at new values for the samples ? |
|
|
Spindrift
Spindrift
Started Topics :
33
Posts :
1560
Posted : Apr 28, 2005 18:27
|
Yes. Basically you need to discard samples to downsample. I don't know how advanced most downsampling rountines is, and there might very well be som interpolation going on, but nevertheless you have to loose those extra samples.
From what I learned it's more to loose than gain by using 48k if you are not planning to release the music on DAT mainly which I guess is not a plan except for the most fanatic underground audiophiles.
Even if the audible difference might be very small there is nothing to be gained from using 48k and just results in hassle.
If you work in 96 it's of course another story since there is a lot more to gain when it comes to getting rid of rounding errors when mixing/processing.
If nothing else it's awkward and more timeconsuming to master and a nuisance for you or any mastering engineer that will have to deal with your masters.
I need to switch my settings on the pulsar to play back the file for example and it tends to bugger out sometimes and play it back in the wrong speed anyway forcing me to reboot.
So I hate 48k for personal reasons.
 (``·.¸(``·.¸(``·.¸¸.·`´)¸.·`´)¸.·`´)
« .....www.ResonantEarth.com..... »
(¸.·`´(¸.·`´(¸.·`´``·.¸)``·.¸)``·.¸)
http://www.myspace.com/spindriftsounds
http://www.myspace.com/resonantearth |
|
|
|
UnderTow
Started Topics :
9
Posts :
1448
Posted : Apr 28, 2005 19:22
|
Spindrift sums it up pretty well. If you are going to go to CD in the end, I would advise 88.2Khz over 96Khz though. 96Khz is good for DVD and broadcast.
Another thing you can do if you have good converters and two digital devices is go through the analogue domain. Best sample rate conversion there is really. 
UnderTow |
|
|
|
DJ Buju
IsraTrance Full Member
Started Topics :
70
Posts :
1334
Posted : Apr 28, 2005 21:13
|
Quote:
|
On 2005-04-28 18:27, Spindrift wrote:
Even if the audible difference might be very small there is nothing to be gained from using 48k and just results in hassle.
If you work in 96 it's of course another story since there is a lot more to gain when it comes to getting rid of rounding errors when mixing/processing.
|
|
Very Very True !
www.domorecords.com
www.myspace.com/domorec
www.myspace.com/tupanrecords |
|
|
|
Hayez
Started Topics :
8
Posts :
393
Posted : Apr 28, 2005 21:48
|
Quote:
|
On 2005-04-28 18:27, Spindrift wrote:
I don't know how advanced most downsampling rountines is, and there might very well be som interpolation going on, but nevertheless you have to loose those extra samples.
|
|
there is a lot of difference in quality between SRC's. check this:
http://www.audioease.com/Pages/BarbaBatch4/Barba4SRCTest.html
"a new art came into my mind which only you can create, the Art of Noises, the logical consequence of your marvelous innovations." Russolo, 1913 |
|
|
|
Spindrift
Spindrift
Started Topics :
33
Posts :
1560
Posted : Apr 28, 2005 23:34
|
Quote:
|
On 2005-04-28 19:22, UnderTow wrote:
If you are going to go to CD in the end, I would advise 88.2Khz over 96Khz though. 96Khz is good for DVD and broadcast.
|
|
You are probably right that it does make sense to use 88.2 for CD since the resampling would be a much easier conversion.
I think the reason most people go for 96k instead is that you will be more futureproof and you8'll have your masters ready for DVD audio.
Of course then you should preferrably use 192k/64bit during production but 96 is nevertheless a little bit more future proof.
But lets leave that discussion for a couple of years...my current computer don't do so well even with 24/96
(``·.¸(``·.¸(``·.¸¸.·`´)¸.·`´)¸.·`´)
« .....www.ResonantEarth.com..... »
(¸.·`´(¸.·`´(¸.·`´``·.¸)``·.¸)``·.¸)
http://www.myspace.com/spindriftsounds
http://www.myspace.com/resonantearth |
|
|
|
DJ_Chorman
IsraTrance Junior Member
Started Topics :
45
Posts :
85
Posted : Apr 29, 2005 02:20
|
why limit yourself though??
you can always go down in quality, never up... |
|
|
|
Boobytrip
IsraTrance Junior Member
Started Topics :
39
Posts :
988
Posted : Apr 29, 2005 15:32
|
|
i... i am speechless... thnx for the answers I guess i'll stop wasting harddiskspace and cpu cycles on 48 KHz then. |
|
|
|
Elad
Tsabeat/Sattel Battle
Started Topics :
158
Posts :
5306
Posted : Apr 29, 2005 15:52
|
i personly wouldnt.
specialy if u dont make audio recordings u can just work 16/44 for easy life and when done export/bounce with 32(24)/96 (or 88.2 if u insist) with correct settings in audio card,then i also save cpu and also get best result (if there is audio recording i record 32 bit - work with 16 bit again and in main export will go back to 32 )
and after proper dithering u will notice the mix is clearer even after converted to 16 bit then just render all 16 bit...
www.sattelbattle.com
http://yoavweinberg.weebly.com/ |
|
|
|
UnderTow
Started Topics :
9
Posts :
1448
Posted : Apr 29, 2005 16:01
|
Quote:
|
On 2005-04-28 23:34, Spindrift wrote:
Of course then you should preferrably use 192k/64bit during production but 96 is nevertheless a little bit more future proof.
|
|
Actually, no. 192Khz is a scam invented by the manufacturers. According to the Nyquist theorem, you need double the sampling rate plus a little bit to sample a finite frequency range. With audio, that is 2x 20Khz + a bit. If you want to be completely sure, lets say that we might hear some harmonics at very low levels up to 30Khz (and this is being extremely cautious). Then you need about 60Khz of sampling rate.
Now you might wonder what harm it might do to go beyond that. More is always better, right? Well in this case it isn't because when you increase speed (sampling rate) you loose accuracy. 192Khz sounds LESS accurate than 96Khz. This is maths and physics. You can't get arround that. Everytime you increase the sampling rate beyond what is absolutely necessary, you have a precision trade-off. (This is provable mathematicaly and has been demonstrated in testing).
96 Khz is allready overkill. The only reason it sounds better sometimes with current technology is because of the anti-aliasing or decimator filter designs that are used.
Instead of just increasing sampling rates, which is relatively easy to do, manufacturers should spend their R&D time and money on improving filter design and costs. But (some) manufacturers have chosen the easy route and are effectively ripping off their customers.
UnderTow
|
|
|
|
Spindrift
Spindrift
Started Topics :
33
Posts :
1560
Posted : Apr 29, 2005 17:49
|
I haven't done any blind tests on 192k quality recordings and can not say anything about it from experience.
I know with converters that higher resolutuion is not equal to better sound and if you spend the same money on a 20bit/44.1k you can many times get a better sounding converter for your money than if you buy a cheap 24/96 converter.
But on my Pulsar you hear a clear difference if you use 96k sampling rate for the synths, but halfs your DSP power unfortunally. Thats a different issue than recording I understand though, and the will still sound better after recording down to 44.1k.
Hearing is not understanding sound according to maths. the maths is applied to explain the hearing among other things.
Take a sine wave at 16 000k. If you have 44.1k that don't give a lot of samples per cycle and it wont look much like a sine any more.
Oversampling will correct it a bit and such high frequencies is hard to tell if it's a triangle or sine wave, but it can help to get that little bit extra sweet sounding treble.
Sure even a 128k mp3 at 22k sound ok for most people and when we talk about this stuff it's minute differences in the end result.
But some people do care about those minute differences and therefore 96k is more futureproof since it will be in the standard of the future.
But it matters very little, especially in this scene since chances are small that one will get the tracks re-released as DVD audio.
And the audible difference is probably very small indeed compared to using 88.1, 96 or 192k.
So in the end you can pick whatever number that is good luck for you really
(``·.¸(``·.¸(``·.¸¸.·`´)¸.·`´)¸.·`´)
« .....www.ResonantEarth.com..... »
(¸.·`´(¸.·`´(¸.·`´``·.¸)``·.¸)``·.¸)
http://www.myspace.com/spindriftsounds
http://www.myspace.com/resonantearth |
|
|
|
UnderTow
Started Topics :
9
Posts :
1448
Posted : Apr 29, 2005 18:35
|
You are right Spindrift. I should have been more precise and explained that I was talking about converters specificaly. (Hence the use of the word sampling throughout my post).
Oversampling within the digital domain is a different thing indeed and can achieve better results although usually this has to do with other factors.
When you say that hearing is not understanding sound according to maths I don't entirely agree. If there is a difference in the way things sound at higher sampling rates, we should find out objectively what is causing this difference. Once we understand the mechanics behind this, we can apply it to anything and everything. Unfortunately I don't think anyone is doing this research due to vested interests.
As far as I know, all the listening tests that have been done so far involve 192Khz ADCs with a decimator to switch to 96Khz. In other words, what is being tested is the decimator and not the sampling rates.
On the other hand, manufacturers would do well to do research into increasing bit-depth. This _does_ give an increase in accuracy and dynamic range. IMO that is where the future lies.
Quote:
|
Take a sine wave at 16 000k. If you have 44.1k that don't give a lot of samples per cycle and it wont look much like a sine any more.
|
|
Actually this isn't correct. Let me explain it with basic maths: To represent a line mathematicaly, you don't need an infinite number of points. You only need TWO points. With those two points you have all the information you need about that line. To represent a circle (or sine waves) you need THREE points. No more, no less. Translated into sampling technology, you only need three samples to represent a sine wave and that can be done up to (nearly) half the sampling frequency.
This is the basis of Nyquist's theorem and it is applied succesfully in alot of different types of technology besides audio. No one has ever disproven his theories and besides the audio world, no one is willfully trading in accuracy for absolutely no gain (higher sampling rates). It just doesn't work.
In other words, with a good ADC/DAC line, a 16Khz sine wave will come through perfectly. There is absolutely no loss or distortion. Any distortion that does occure has to do with problems in electrical engineering design. Not with the maths or physics involved. Hence my comment about manufacturers spending their research money and time in the right direction.
I hope this clarifies my views a bit.
UnderTow
|
|
|
|
Spindrift
Spindrift
Started Topics :
33
Posts :
1560
Posted : Apr 29, 2005 19:11
|
Sorry, but I still can't grasp how a 16k signal can be a sine when sampled less than three times per cycle.
If we talk about a sine at 20k sampled with the frequency of 40k you get two samples per cycle, which means that you cannot tell if you have a sine or a triangular waveform from the resulting information.
I just can't grasp how it would be possible.
A line can be represented from knowing only two points, but a curved line cannot and you need more variables to calculate a sine or a higher resolution when you sample point by point.
I'm programming a little racing game right now and wish that I only needed two values to handle the car in the curve, but unfortunally not since sines is a bit more complicated mathematically.
Nyquist is of course right that you need double the sample frequency of the frequency you want to represent.
In most areas except hifi audio use the frequency is all that matters and the waveform is of not so much importance.
But I don't really know how it works with say a square or saw wave at very high requencies since you need overtones to have reshape a sine normally.
So since there is not any room technically for octaves of the sound everything high up would be represented as sine waves or rather triangular waves.
To have a real saw at 20k you would have to have a very high samplerate, but probably our hearing can't tell the difference really anyway so it's very theoretical I would say.
(``·.¸(``·.¸(``·.¸¸.·`´)¸.·`´)¸.·`´)
« .....www.ResonantEarth.com..... »
(¸.·`´(¸.·`´(¸.·`´``·.¸)``·.¸)``·.¸)
http://www.myspace.com/spindriftsounds
http://www.myspace.com/resonantearth |
|
|
|
fregle
IsraTrance Junior Member
Started Topics :
11
Posts :
982
Posted : Apr 29, 2005 19:11
|
undertow, spindrift: that was very interesting, tnx...
clear as freshly pumped up water undertow... |
|
|
|
UnderTow
Started Topics :
9
Posts :
1448
Posted : Apr 29, 2005 20:49
|
Quote:
|
On 2005-04-29 19:11, Spindrift wrote:
Sorry, but I still can't grasp how a 16k signal can be a sine when sampled less than three times per cycle.
If we talk about a sine at 20k sampled with the frequency of 40k you get two samples per cycle, which means that you cannot tell if you have a sine or a triangular waveform from the resulting information.
I just can't grasp how it would be possible.
|
|
That is why you always need a little bit more than twice the maximum sampled frequency because you need 3 points to define a sine wave (or circle). That is why 44.1Khz has a maximum sampled frequency slightly below 22.05Khz.
Quote:
|
Nyquist is of course right that you need double the sample frequency of the frequency you want to represent.
In most areas except hifi audio use the frequency is all that matters and the waveform is of not so much importance.
|
|
Actually, audio needs to be less accurate in many ways than many other systems. I am talking about data communication, satellite links, laser systems, weighing systems, medical systems, military systems etc ... We can do alot to audio before people complain. Hence 128Kbps mp3s. 
Quote:
|
But I don't really know how it works with say a square or saw wave at very high requencies since you need overtones to have reshape a sine normally.
|
|
That is where Mr Fourrier comes in. It is all sine waves in the end.
Quote:
|
So since there is not any room technically for octaves of the sound everything high up would be represented as sine waves or rather triangular waves.
|
|
No, sinewaves. A square wave or triangle wave at 16Khz is composed of a fundamental sine wave and harmonics above it. The harmonics that fall below 20Khz are still sinewaves and the one's above (still sine waves) we can't hear so they are pointless and thus don't need to be sampled.
Quote:
|
To have a real saw at 20k you would have to have a very high samplerate, but probably our hearing can't tell the difference really anyway so it's very theoretical I would say.
|
|
Indeed. We can't hear the harmonics of a saw wave at 20Khz so it is irrelevant for audio. We can hardly hear (if at all for some people) the sine at 20Khz anwyay.
UnderTow |
|
|
|
