Ask an Audio geek

[Alex K]

This is an example I found of what frequency limitation does to a sharp pulse:



Not a perfect illustration but the best I found. Notice that the resulting filter response (red) has these typical broad movements which will be picked up if you sample at twice its critical frequency, whike the hard transients are gone. This is not an artefact of bad electronics, that's what it means to limit the frequency of the signal. A turntable cartridge stylus would do the same thing to short pulses due to it mechanical inertia. Our ears do the same thing. If you don't hear or transmit above 20k, you cannot physically distinguish between that pulse and the response curve - they become the same (up to phase shifts near the cutoff frequency which depend on the type and steepness of filter used, but these effects can be reduced arbitrarily by choosing a higher sample rate if they bother you. But they are also present in any analog gear and your ears. Still, one might be motivated to sample at 96k in order to be removed two octaves from the hearing range. That way, one can use a given filter and more perfectly avoid aliasing effects and such, and other weird cutoff effects like phase shifts will be even further removed from the hearing range)

No no, it has nothing to do with the psychoacoustic effect (that's relevant for dropping in-between frequencies in mp3 compression), it has to do with the limited frequency range of hearing.

The point of the sampling theorem is that with a signal of limited highest frequency, these discrete points uniquely determine what goes on between them. If the signal you send in is cut off sharply at say 20 khz using a steep filter, it will not have any features which are so quick that they can be missed if you only measure 44100 times per second. Such a signal can only change its direction of up-down movement so fast, because changing faster would require containing higher frequencies. If you have ever seen what a low pass filter does to short pulses, it might become clearer.

[AFTT47]

Yeah, I don't think DCC ever made it market though. I know mini-disc did because I was one of the boneheads that bought one. It was a walkman-like device. Then there were the DAT decks but I believe those were lossless like CD. I think the convenience factor (or lack thereof) killed DAT.

[Alex K]

DAT was/is 16bit 48000 khz lossless afaik, so a little bit better than CD. They had their niche, I almost bought one. I once borrowed an MD device from a teacher for a simple live recording I had to do, but was never tempted to buy one because by the time I would have considered it, I had a discman that could read burnt CDs.

[ignoramus]

Guys, did DAT use the same principles of recording like CD?
I know they were used professionally for a few years...

[AFTT47]

Guys, did DAT use the same principles of recording like CD?
I know they were used professionally for a few years...

Same principles, slightly higher specs. It made sense for a professional format where conveniences like instant skip ahead/back were not important. Lack of such probably doomed it as a consumer format.

[Alex K]

Since it is very hard to build a filter circuit which effectively shuts off the signal at 22 khz while still leaving it untouched at 20khz, and the ones you can build from ordinary analogue circuitry have strange properties, it is better to use an AD-Converter which operates at a much higher sample frequency and use a more moderate conventional filter on the input side. When converting back down to only 44100 samples per second via number crunching in order to press a CD, one can use much more precise computer-based digital filtering operations which can cut frequency very steeply without producing weird phase shifts. That way, one can basically avoid aliasing without having to use a conventional filter circuit with insane steepness which ruins sound quality.

Aliasing, if you haven't heard it, is something you know from movies - when a spoke wheel turns fast enough on film that the next spoke replaces the last one from one frame to the next, it will look as though the wheel stands still. When it turns even faster, it will look as though it turns slowly backwards. The same happens with audio sampling. If you sample at 44100 times a second, a signal with 30000 Hertz will result in a fake recorded signal of 16000 Hertz, in the hearing range. This is why one needs to very strictly filter high frequencies out before sampling. If you don't do it, ultrasound gets converted to horrible noise, and if you do it badly, like in some early digital recordings, it sounds like crappe because the filters distort the signal in the hearing range. Again, with digital filter algorithms working with high precision numbers, maybe even floating point, this is not an issue any more.

The only objection one might have against a perfectly realized 44.1 digital recording, apart from quantization noise due to the 16 bits, is that it really brickwall-filters everything above 20khz or so, whereas any analogue equipment will roll off more smoothly. I don't know whether that might change perception of the sound, but I doubt it.
In any case, it is something one can arbitrarily improve by putting higher sample rates on the medium. If we go to 96k samples per second as is already done in many places, the hard cutoff is ideally at 48 kHz, so far away from the hearing range that I strongly doubt that there is a chance in hell to hear the difference between a slow rolloff starting near 20kHz or a hard cut near 48kHz.

But there definitely are enough things that can go and have gone wrong to let a CD sound objectively worse than vinyl in ways that are qualitatively different from the imperfections one gets from analogue recordings, and it needed a lot of sophisticated technology improvements and digital signal processing techniques like those I mention above, to really reach near the mathematical ideal.

[Alex K]

On dithering:

https://www.youtube.com/watch?v=IRlohQw-1DY

[paulpablo]

normalize, compress, add reverb.  What is the best order to do these things in?  Is there a best order to do these things in?  Does it depend on the circumstances?

[Alex K]

It doesn't make sense to normalize before the final step because all those other things are going to change the level.
The order of compression and reverb is an interesting question. I am not an expert at all, but it is fun guessing. I'd love to hear an expert answer.

My guess is that if you add reverb after compressing, you might get a really strange artificial sounding mush because things which originally had different volumes will simultaneously reverb at the same strength. For example, if the compressor dials up the gain going from a loud passage to a silent one, something very loud from before will have the same amplitude of reverb in the final signal as the small sound that comes after, and that seems perverse.

So my pick would be: 1. reverb 2. compress 3. normalize

But if the reverb is not for the overall product but one channel, for example a mic picking up a singing voice, it might make more sense to reverse and do the reverb after compression to have it overall sound more uniform. Again, pure guessing here.

[Alex K]

reverb, compress, normalize is also the order in which a natural room reverb will come into the mix, so it seems to my naive self like the natural thing to do it the same way artifically to an entire recording if it sounds to dry.

Again, to me it might not make sense on an individual mic part of a larger performance because what will then happen is that if you put reverb before compression on that channel, whenever the singer is silent, the more and more silent tail of the reverb gets increasingly blown up by the compressor. If that is your desired effect, fine, but it might sound bad in some instances.