Why 24 bit audio and anything over 48k is not only worthless, but bad for music.
post-11816088

#### Gr8Desire

What do you mean "induce" dithering. You either add dither or you don't. The point of it is to randomize the errors due to quantization, substituting broadband noise for the truncation distortion due to the rounding. Quantization intimately relates to the bit depth, not the sample rate, as the number of bits determines the numerical set onto which we must map the samples.

I am using the term from my experience in writing related algorithms.  Inducing dither is correct because that's what the algorithm does.

When you change timebases you have to estimate the quantized values for the new timebase. Dithering (which is the same term used in image processing where I have more experience) is induced into the result to help "guess" at missing data points (or as you might say: quantized values in a different non-integral timebase).

FWIW: Audiophiles have little actual experience, so they get these simple ideas things wrong.

And as I said before, you don't dither when changing with bit depths. No need to. Any missing values can be accurately determined by interpolation when up sampling or just removing data points when down sampling.

post-11816172

#### icebear

I've tried the book, the baby, the bathwater, the bathtub, I even tried being nice once(I didn't like it), nothing seems to work. ...
... there is still the kitchen sink, you have forgotten that might finally work

post-11816226

#### icebear

2^16=            65,536
2^24=     16,777,216
2^32=4,294,967,296

2^8= 256

With 16 bits data can represent 65,536 different volume levels and the signal is sampled 44100 times per second. The theoretical dynamic range capacity is way enough to accurately capture actual music. A couple of examples exist that you can cut down the word length to 8 bits where quality loss gets pretty obvious but stil at 10 or 12bits, depending on the music type it's rather difficult to tell degradation right away.

I am not sure which marketing genius used the jagged steps picture for the first time in the early 80's, most likely someone of Philips or Sony. There are no steps, there are isolated points that get connected when the signal is coverted to a continous analog sine type wave, all "steps" are GONE. And 44100 data points per second ... they are already pretty tightly packed next to each other
.

Enjoy the music and don't worry about the numbers folks. If it sound right to your ears, if the music involves you emotionally and is fun to listen to and you forget about the equipment, then it's good enough. No matter what numbers.

post-11816423

#### cjl

I am using the term from my experience in writing related algorithms.  Inducing dither is correct because that's what the algorithm does.

When you change timebases you have to estimate the quantized values for the new timebase. Dithering (which is the same term used in image processing where I have more experience) is induced into the result to help "guess" at missing data points (or as you might say: quantized values in a different non-integral timebase).

FWIW: Audiophiles have little actual experience, so they get these simple ideas things wrong.

And as I said before, you don't dither when changing with bit depths. No need to. Any missing values can be accurately determined by interpolation when up sampling or just removing data points when down sampling.
Dither can also be used for reduction in bit depth though (and it frequently is). If you simply truncate or round each value, you get quantization noise, which is correlated with the signal and generally undesirable (though inaudible for 16 bit, at least in every case I can think of). If you add dither when reducing the bit rate, quantization noise can be eliminated in favor of a broadband noise completely decorrelated from the signal, and of any character you want (usually shaped to minimize audibility).

post-11816463

#### RRod

I am using the term from my experience in writing related algorithms.  Inducing dither is correct because that's what the algorithm does.

When you change timebases you have to estimate the quantized values for the new timebase. Dithering (which is the same term used in image processing where I have more experience) is induced into the result to help "guess" at missing data points (or as you might say: quantized values in a different non-integral timebase).

FWIW: Audiophiles have little actual experience, so they get these simple ideas things wrong.

And as I said before, you don't dither when changing with bit depths. No need to. Any missing values can be accurately determined by interpolation when up sampling or just removing data points when down sampling.

What do you mean "removing data points" when changing bit depths. If you change from 16/44.1 to 8/44.1, you have just as many samples, but the values of those samples must be mapped to fewer integers.

I see what you mean about sample rate conversion, but the dithering there still seems ancillary to doing calculations at higher bit depth and then bringing things back down.

post-11816477

#### RRod

Dither can also be used for reduction in bit depth though (and it frequently is). If you simply truncate or round each value, you get quantization noise, which is correlated with the signal and generally undesirable (though inaudible for 16 bit, at least in every case I can think of). If you add dither when reducing the bit rate, quantization noise can be eliminated in favor of a broadband noise completely decorrelated from the signal, and of any character you want (usually shaped to minimize audibility).

And when people talk about dither (and "hearing it") across the boards here, they mean exactly this. The feel that the use of it when quantizing from 24 or 32-bit down to 16-bit produces audible artifacts that both a) change in quality when different dithering algorithms are used and b) make 16-bit delivery sound "clearly" inferior to 24-bit. They would say this even for 16/44.1 vs. 24/44.1 where no sample rate conversion is needed. Now whether we're all using the term incorrectly, I'm down for discussing that.

post-11816645

#### jcx

any example of people doing this with rational gain structure, better perceptual noise shaped dithers - not just turning up the volume when the music is low to hear the dither?

post-11816754

#### castleofargh

##### Sound Science Forum Moderator
just to be clear on something, are all the "I'm an idiot" or "I have no idea what I'm doing" situations, a part of what we have to consider as the norm? or can we just presuppose that when we talk about a case, we talk about the people using the devices and software as they should be paired and used?

because if we take in the missuses as if they're standard uses, then I have to say I believe in cable sound, I believe that amps all can sound different. and of course after using -40db gain on my tracks and using windows volume at 2% and compensating with the amp maxed out, I believe that different dither choices might just become audible among other things.

post-11816877

#### interpolate

8-bit is going sound like a video game or a glitch techno drumbeat. It in theory does not have a large storage array, so it's kind of pointless comparing 16-bit to much lower bit-depths don't you think.

Dither algorithms are designed to make use of the frequency attenuation as we all know. In a way, everybody has become very complacent and regards that everyone should be knowledgeable about everything under the sun. The only idiots are those are presume this and denounce others for not.

post-11816921

#### RRod

8-bit is going sound like a video game or a glitch techno drumbeat. It in theory does not have a large storage array, so it's kind of pointless comparing 16-bit to much lower bit-depths don't you think.

Dither algorithms are designed to make use of the frequency attenuation as we all know. In a way, everybody has become very complacent and regards that everyone should be knowledgeable about everything under the sun. The only idiots are those are presume this and denounce others for not.

Here's a 5sec clip of 8bit versus 16bit, from the kind of track where you can get away with it. The 8bit version was made by truncation (no dither), and then zeros were padded back on to put it back into 16/44.1 format. Enjoy:

post-11816926

#### castleofargh

##### Sound Science Forum Moderator
8-bit is going sound like a video game or a glitch techno drumbeat. It in theory does not have a large storage array, so it's kind of pointless comparing 16-bit to much lower bit-depths don't you think.

Dither algorithms are designed to make use of the frequency attenuation as we all know. In a way, everybody has become very complacent and regards that everyone should be knowledgeable about everything under the sun. The only idiots are those are presume this and denounce others for not.
notice that I'm wording 2 options with the use of "or" in the middle. I'm not saying everybody is an idiot, most just don't know enough to understand that they don't know and fall into my second option. like you when you mistake 8bit music from old video games for the actual sound of 8bit music

and I still believe my question to be legitimate. are we talking about correct use of thing? or do we need to talk audio like we talk micro waves and put warnings all over it to tell people not to dry their pets with it? and take the exploding yorkshire as stuff that happen to some people? or can we limit our talks to the proper use of a microwave? that really leads to a very different debate IMO.
and I never ever thought that someone not knowing was similar to someone being an idiot. never ever. we all start somewhere with no information and go somewhere. still idiots do exist!

post-11816929

#### interpolate

Yeh OK. But what does that prove?

8-bit audio was the norm for years in computing however up until 16-bit, the only real choice was analogue.

post-11816937

#### RRod

Yeh OK. But what does that prove?

8-bit audio was the norm for years in computing however up until 16-bit, the only real choice was analogue.

Just proves that 8bit doesn't have to sound like a video game. But since there are genres that are not sludge metal, you need more bits to get the dynamic range you need. I've yet to find anything that really needs more than 14, at least listening on my HD800 setup. With really well sealed IEMs, maybe I'd find something that needs a 15th bit.

post-11816941

#### interpolate

OK. Granted.

I'm much more a producer than a listener, so do have an interest in such things although with modern audio anything below 16-bit isn't really worth caring about. In Low-Frequency filters and bit-depth decimation it becomes more useful for effects and in some cases DSP effects where inaudible audio is enhanced at a fundamental level.

post-11816960