quote:

Originally posted by DigiNut You're still propagating the same patently false nonsense about losing bits though. It just doesn't work that way. Good digital devices never use any fewer bits when attenuating the signal, they just change the reference level that corresponds to the signal maximum. Before the DAC, you change the signal reference. After the DAC, you alter an op amp.

I'm talking integer here as I'm currently aware of no output device that supports float output. Integer does not have any reference or exponent. Take a 16bit fullscale sine, devide by to, thus truncating the LSB, multiply by two to restore original level. The LSB might be stuffed with zeroes or dither noise, in any case although your data has never left 16bit integer domain the information contained is left at only 15 bits.

Digitally controlling the output gain of the DAC is not what I understand as digitally amplifying/attenuating... and it's not the way most players do their output attenuation (winamp might, but i don't know)

quote:

You're also wrong about roundoff error. Integer division might introduce an error, but not necessarily. It's only in floating-point math where you might multiply two perfectly accurate numbers and end up with an inaccurate one ("0.4" is not perfectly representable, for example). In practice, though, the error is so small that it won't make a difference. 0.4 comes out as 0.400000000002 or something like that.

That's all I said, I was talking of a multiple feedback scenario.
...and your contradicting yourself

btw: integer division does nearly always introduce error. As it's rather uncommon that you devide by a common factor.

quote:

Converting from floating point back to integer is an issue, and that's what the dithering algorithms in software are for. UV22HR does a great job of making the change almost completely transparent. Waves IDR does it pretty well too.

dither does not get rid of errors it's efficient in masking them ...by introducing even more errors (also known as noise).

___________________

DISCLAIMER:
The above text is believed to be derived from objective logic. No personal offence, implicit or explicit, is intended. Ambiguous expressions are to be interpreted in the less offensive way.

quote:

Originally posted by halo I'm talking integer here as I'm currently aware of no output device that supports float output. Integer does not have any reference or exponent. Take a 16bit fullscale sine, devide by to, thus truncating the LSB, multiply by two to restore original level. The LSB might be stuffed with zeroes or dither noise, in any case although your data has never left 16bit integer domain the information contained is left at only 15 bits.

You're confusing reference with mantissa. What I'm talking about is using one 16-bit word (or two, in stereo) to represent the signal, and another 16-bit word internally to represent the swing. It's still a 16-bit signal on the input and output, but there's more to it than that inside the device. I can't say for sure that every digital device does this, but doing it that way will never run into the scenario you speak of. Software always uses floating-point internally anyway, which is the "digital" that most people here are talking about, so the integer division issue is a moot point. And since most VAs today are just software on the inside, they have the same characteristics.

I should also point out that the actual level of "noise" we're talking about here (i.e. miniscule), if one accepts that it even exists, is no larger than the THD characteristics, line noise, and other nonlinearities introduced by the op-amps in analog devices. Worse, in the analog domain, you don't even have to apply any gain - even with unity gain you'll still be adding noise.

It's very, very rare that 24-bit or even 16-bit DSP would introduce more noise than doing it in the analog domain. I've never seen it happen in practice and I've never seen a good theoretical explanation either.

quote:

Digitally controlling the output gain of the DAC is not what I understand as digitally amplifying/attenuating... and it's not the way most players do their output attenuation (winamp might, but i don't know)

I'm not sure exactly how the internals of WMP or Winamp work, but I know that what you're saying is not correct in this instance because I can turn the volume on either one of the down to just one notch above the minimum, and still hear zero distortion on the output. Again, by your logic this would be limiting the signal to 1 or 2 bits. Obviously this would be heavily distorted, and obviously, this is not the case.

The master volume on a sound card usually controls an op-amp after the DAC in the signal path. I know this is not a software function because it's not available for every audio interface. There's definitely no loss there, and however Windows/WMP/Winamp/anything does it, it's not truncating any bits. At the risk of being repetitive here, it's very easy to prove this by turning the volume all the way down to the minimum and cranking your master volume to compensate. There's no distortion, and there should be if we were truncating more than half the bits.

quote:

...and your contradicting yourself

How so?

quote:

btw: integer division does nearly always introduce error. As it's rather uncommon that you devide by a common factor.

Again, where's your proof of this?

quote:

dither does not get rid of errors it's efficient in masking them ...by introducing even more errors (also known as noise).

That's true... more or less. It doesn't introduce "even more" errors, it introduces a different kind of noise, but it's the exact same amount of noise as truncation. In any event, I never said that dithering gets rid of errors. I just said that's what used when converting floating-point signals back to integer signals, and if you're only ever doing that once in your signal path (which is usually the case), the noise will never be audible with a good dithering algorithm.

___________________
My party schedule:
2009-02-21 - DJ Attention @ I'm So Popular
2009-06-18 - DJ Annoying @ People Need To Know Where I'll Be
2012-11-32 - DJ Insufferable ɸ Or At Least the Stalkers I Complain About
2048-06-66 - Spastic & Whocares ¶ Although I'm Actually Flattered
9999-45-81 - Tweaker Gimp ☼ I Probably Won't Even Go To This But I Have To Make Sure I Fill Up All The Available Space Here

blah blah blah blah......
is all I hear hahaha jk what are you guys talking about!!!!!

___________________
myspace.com/jmancha
irc.gamesurge.net #team-rush

quote:

Originally posted by DigiNut I've heard this bizarre assertion about "bit loss" several times and still can't understand where it could possibly have come from. It's completely untrue, and my only guess is that it must have been something spewed by a marketing weasel that got misinterpreted by a few viewers or buyers. Bit depth in the digital domain is a constant 16-bit, 24-bit, or 32-bit floating-point. Period. There are no bits "removed" when you lower the volume, nor is there ever a scenario where you "don't get the full 16/32 bits" (this is not like a motor assembly where you only get 90% efficiency, you get 100% of the swing in a digital signal). Only exception is when you're using a bitcrusher, but that's obviously a deliberate bit reduction (duh). In actual fact, it's much better to use software to reduce gain because there's no noise. In the analog domain, if you have a noisy stage that outputs low gain, and you amplify that at the next stage, you're amplifying the noise with it. This is a non-issue when working with digital audio.

I think I know where this issue came from; when you record audio into the digital domain, you should attempt to do it at the highest level possible, in order to make the best use of your bit depth. try recording something at low volume then boost up the volume (say in ableton live) and you can see on the waveform that it is very sort of "quantised" looking almost, sort of jagged. Also, Remember that it is common practise to dither your work when bouncing down, I don't know if that means that this also applies to output... either way, I doubt whether it makes a significant difference! At the end of the day, monitoring can be done quite well with very little, Chicane did the whole of behind the sun on a pair of monitors powered by a teac amplifier, he did quite well from that album I believe...

Very interesting thread, but there's so much contrasting information.

I can't really add to it, because my output is turned halfway down in windows, and halfway down on my monitors... now I'm actually doubting the way I'm handling my chain.

I'm going to play around with various settings and see if I can actually hear any audible difference in sound quality.

quote:

Originally posted by kitphillips I think I know where this issue came from; when you record audio into the digital domain, you should attempt to do it at the highest level possible, in order to make the best use of your bit depth. try recording something at low volume then boost up the volume (say in ableton live) and you can see on the waveform that it is very sort of "quantised" looking almost, sort of jagged. Also, Remember that it is common practise to dither your work when bouncing down, I don't know if that means that this also applies to output... either way, I doubt whether it makes a significant difference! At the end of the day, monitoring can be done quite well with very little, Chicane did the whole of behind the sun on a pair of monitors powered by a teac amplifier, he did quite well from that album I believe...

Sort of but not quite there.

If you are taking any analogue signal and converting it to digital (i.e. recording from a mic/guitar to a soundcard input) you want to record to the highest bit depth and samplerate you possibly can.

Increasing gain has no effect on quantisation. When you record a sound and play it back in a wave editor increasing gain just increases the vertical height of the waveform as this is just a graph of amplitude over time. Its frequency is represented in the number of full oscillations per second. So it can make a wave sound appear more 'jagged' but only because you are stretching it vertically.

On the issue of pre gain and post gain this is to do with mic/instrument and line level devices. When you record from a guitar to a line level soundcard input there is an impedance difference and a peak level difference. I forget what the impedance difference is but its something on the order of 10,000 ohms I think. Don't quote me on that. The peak level difference is on the order of 40 to 60 decibels.

If you record a mic or a guitar straight into a line level input with no preamp stage, you have to apply about 40 to 60 decibels of post gain to the recording - which will increase the noisefloor in proportion to the peak signal - basically there will be lots of hiss and low level noise but its now 40 to 60 decibels louder so you can hear it really badly. If you don't apply 40 to 60 decibels of post gain then you are just left with a recording so quiet that it is for all intents and purposes unusable when composited with other sounds recorded from line level devices.

On the issue of dither. You don't dither when you bounce. You only ever dither when you are bouncing down to a lower bitdepth. If you are recording at 32 bit float and you bounce to 32 bit float you should not dither.

However, if you record at 32 bit float and bounce down to 16 bit you will need to dither but you do not want to dither more than once as this process involves adding low level noise to a recording. This noise is cumulative. This is why the most common advice you will get about dither is to work at the highest bitdepth you can for as long as you can and then when everything is done, you dither down to CD Audio bitdepth as the very last process. But only if that mix is destined for 16 bit CD.

quote:

Originally posted by DigiNut You're confusing reference with mantissa. What I'm talking about is using one 16-bit word (or two, in stereo) to represent the signal, and another 16-bit word internally to represent the swing. It's still a 16-bit signal on the input and output, but there's more to it than that inside the device.

I'm not... mantissa is the only representation kind of like a reference i could think of in device independant digital domain. Of course soundcards might store their gain values in separate integers and whatnot but as i pointed out this is not digital attenuation.

quote:

I can't say for sure that every digital device does this, but doing it that way will never run into the scenario you speak of.

Absoluely right in any case all you're using to attenuate is the control panel of your soundcard.

quote:

Software always uses floating-point internally anyway, which is the "digital" that most people here are talking about, so the integer division issue is a moot point.And since most VAs today are just software on the inside, they have the same characteristics.

Absolutely right. As long as the signal is kept inside one and the same digital environment.

quote:

I should also point out that the actual level of "noise" we're talking about here (i.e. miniscule), if one accepts that it even exists, is no larger than the THD characteristics, line noise, and other nonlinearities introduced by the op-amps in analog devices. Worse, in the analog domain, you don't even have to apply any gain - even with unity gain you'll still be adding noise. It's very, very rare that 24-bit or even 16-bit DSP would introduce more noise than doing it in the analog domain. I've never seen it happen in practice and I've never seen a good theoretical explanation either.

You're right from the practical view... Professional analog equipment should not humm and will not if set up right. but that does not mean that there is no risk of humming. Now plugins in a digital environment are way more forgiving towards a sluggish configuration but that does not mean there is no potential for noise.

quote:

I'm not sure exactly how the internals of WMP or Winamp work, but I know that what you're saying is not correct in this instance because I can turn the volume on either one of the down to just one notch above the minimum, and still hear zero distortion on the output. Again, by your logic this would be limiting the signal to 1 or 2 bits. Obviously this would be heavily distorted, and obviously, this is not the case.

WMP uses the fader of the wave output channel in the media contol panel. So rightly this will normally push volume handling to the soundcard. As i said, i don't know about winamp, but i doubted, as i have never seen the controls in the media panel move when contolling winamps volume. but that was back in the days
Let's take any sequencer (i will refer to logic) that renders a single output stream to the media driver. Let's use the standard windows 16bit sound driver. (ASIO can be rendered to in 32bit float and the driver might be intelligent enough to rescale and adapt gain before rendering to the DAC) let's say we normalize the output to -6dB for headroom reasons will loose the one bit the moment we render to output.
Now for part of being audible... how prominent noise will be depends strongly on the type of signal you're using. Remember that subbands in the MPEG data stream may be coded in 2 or less bits without introducing _audible_ noise.
As you correctly said digital noise in the LSB will be shadowed by the noise of the analog outboard so you will be unable to hear that unless you (digitally) amplify before you convert to analog. In any case you will either loose bits or reduce analog SNR.
On the other hand, 5bit still give you a SNR of 24dB let's say your last bit is at 0dB(A) (so 16bit FullScale is 96dB(A)) a signal in the least 5 bits will produce an overall level equivalent to the environmental sound in a quiet bedroom. Of course you cannot hear the LSB noise...

quote:

How so?

don't take me serious on that one but...

quote:

Integer division might introduce an error, but not necessarily. It's only in floating-point math where you might multiply two perfectly accurate numbers and end up with an inaccurate one

to me contradicts the

quote:

You're also wrong about roundoff error.

quote:

Again, where's your proof of this?

2/4
45/2
45/3
45/4
... basically integer division by any number that is not a multiple of the number you devide by. That is every other number for division by 2, 2 out of 3 numbers for division by 3...

quote:

That's true... more or less. It doesn't introduce "even more" errors, it introduces a different kind of noise, but it's the exact same amount of noise as truncation. In any event, I never said that dithering gets rid of errors. I just said that's what used when converting floating-point signals back to integer signals, and if you're only ever doing that once in your signal path (which is usually the case), the noise will never be audible with a good dithering algorithm.

you're right... more or less ..as long as you can swear to god that none of your components does any conversion... EVER... i for myself wouldn't count on that... an i do reuse sampled data.

___________________

DISCLAIMER:
The above text is believed to be derived from objective logic. No personal offence, implicit or explicit, is intended. Ambiguous expressions are to be interpreted in the less offensive way.

To make my point clear...

1. usually it's hard to hear digital noise
2. bit loss does happen
3. it might not be obvious in your configuration
4. you can demonstrate bit noise by simple integer division... that's what bitcrushers do
5. you'll have to push your gear to make it audible, but it's only two clicks and a twist to get as far away from digital noise as you can.
6. "...but it normally works" is no excuse for bad design...(see Microsoft)

last but not least: in a normal home listening environment 12 bit (about 70dB) SNR is way more than your dynamics should be. That would be about 100dB peak amplitude over a quite moderate backgound noise of 30dB (most computers push background noise to about 40dB)

___________________

DISCLAIMER:
The above text is believed to be derived from objective logic. No personal offence, implicit or explicit, is intended. Ambiguous expressions are to be interpreted in the less offensive way.

quote:

Originally posted by Derivative Sort of but not quite there. If you are taking any analogue signal and converting it to digital (i.e. recording from a mic/guitar to a soundcard input) you want to record to the highest bit depth and samplerate you possibly can. Increasing gain has no effect on quantisation. When you record a sound and play it back in a wave editor increasing gain just increases the vertical height of the waveform as this is just a graph of amplitude over time. Its frequency is represented in the number of full oscillations per second. So it can make a wave sound appear more 'jagged' but only because you are stretching it vertically. On the issue of pre gain and post gain this is to do with mic/instrument and line level devices. When you record from a guitar to a line level soundcard input there is an impedance difference and a peak level difference. I forget what the impedance difference is but its something on the order of 10,000 ohms I think. Don't quote me on that. The peak level difference is on the order of 40 to 60 decibels. If you record a mic or a guitar straight into a line level input with no preamp stage, you have to apply about 40 to 60 decibels of post gain to the recording - which will increase the noisefloor in proportion to the peak signal - basically there will be lots of hiss and low level noise but its now 40 to 60 decibels louder so you can hear it really badly. If you don't apply 40 to 60 decibels of post gain then you are just left with a recording so quiet that it is for all intents and purposes unusable when composited with other sounds recorded from line level devices. On the issue of dither. You don't dither when you bounce. You only ever dither when you are bouncing down to a lower bitdepth. If you are recording at 32 bit float and you bounce to 32 bit float you should not dither. However, if you record at 32 bit float and bounce down to 16 bit you will need to dither but you do not want to dither more than once as this process involves adding low level noise to a recording. This noise is cumulative. This is why the most common advice you will get about dither is to work at the highest bitdepth you can for as long as you can and then when everything is done, you dither down to CD Audio bitdepth as the very last process. But only if that mix is destined for 16 bit CD.

Sorry, maybe I didn't express myself clearly.. I'm not really refering to hi impedance vs low impedance, what I'm talking about is, when you record a (for example) guitar at a certain volume, you then may have to boost it up a little, do this enough and you see jagged bits (dunno what to call them) if you record the same guitar, at the level you need it at (therefore not boosting it in the daw) you don't see those same jagged bits, despite the fact that the waveform is the same size. I think it has to do with just making the best use of your headroom you can... Anyway, noise floor is a good enough reason alone.

Thanks for the info about dither, I thought that might be the case but I always dithered at the end anyway, now (as I only use 16-44.1 anyway) I know I don't have to bother Although, if this is the case, why does the waves L3 multimaximiser(ducking head from flames) use dithering? Should I turn this bit off perhaps?

I don't know why certain Waves plugins have the option to apply dither on the output. I'm talking about the L series limiters and the Linear Phase Equalisers. The dither won't be applied until you bounce the output so for the love of god turn dither off if it is enabled. The only time you want dither on these plugins is if you have a finished, mastered tune and L3 is like the last post processor in the signal chain.

Also, if you use for instance L3 and Linear Phase EQ in your mixes with dither on, then slam your finished mix through Izotope Ozone or some other mastering app with dither on, this will add at least 3 times the amount of dither noise to your mix which is horrendous.

Its best not to bother with dither until you mixdown is spot on and you are a dab hand at mastering recordings. I don't even concern myself with dither and always turn it off where it is available.

I'm not even going to try to understand anything you guys just said. What was the final verdict? Does reducing the volume in your ASIO mixer software reduce the quality comming out the speakers or not?

Thanks

Short and Sweet: Yes.