quote:

Originally posted by camsr I use my Grado cans straight from the soundcard, so I HAVE to turn down the software mixer level to listen at a reasonable level. But since the slider for the master level is linear, I have it usually set at one-quarter of its full range. 1/4 of a 0dB signal is -12db. This means I only lose two bits of resolution at 16 bits (onboard sound). 14 bits is 2^14 = 16384 discreet volumes for monitoring, OR 84 dB of dynamic range. It's 84 db of range because each bit corresponds to 6dB of headroom. The point quantization noise really becomes super noticeable is at about 12 bits. So you might not be losing as much headroom as you think using the software mixer. The only time when bitdepth is important is when you are processing with plugins. Phase shifts and the like will sound bad at low bitdepths. High bitdepths in the processing stage allows you more headroom for the effects to work without raising quantization noise.

I don't get this reply, we've just been told quantitization/bitreduction does not concur when altering the volume within software as volume levels are relative. :S

You can have 16bit audio with the max output set to 0dB, or 16bits between -inf and -30dB if you lower your volume.

___________________

Storyteller Website | Storyteller @ Facebook | Storyteller @ Beatport | Storyteller @ Soundcloud | Stephen J. Kroos - Europa (Storyteller Remix)
Anthony Mea - Get It On (Storyteller Remix)

quote:

If less is more think about how much more more would be. -Frasier

quote:

Originally posted by Storyteller I don't get this reply, we've just been told quantitization/bitreduction does not concur when altering the volume within software as volume levels are relative. :S You can have 16bit audio with the max output set to 0dB, or 16bits between -inf and -30dB if you lower your volume.

Listen, there's only TWO ways the soundcard is gonna lower the volume:

Digitally controlling a solid-state amp circuit on the soundcards output circuit, AFTER the D/A conversion. It would be controlled by software drivers interfacing with the card.

OR

The digital signal that comes into the soundcard's D/A convertor is attenuated. Raising the volume after this part increases quantization noise.

Decibel measurements ARE relative. But real voltage is not. If you make up for digital attenuation anywhere in the signal chain, it will increase the noise, always.


Not that it really fucking matters or anything

It does matter. Really glad you said that. Things are making more sense each post . Getting smarter day by day :P

___________________

Storyteller Website | Storyteller @ Facebook | Storyteller @ Beatport | Storyteller @ Soundcloud | Stephen J. Kroos - Europa (Storyteller Remix)
Anthony Mea - Get It On (Storyteller Remix)

quote:

If less is more think about how much more more would be. -Frasier

This is more how I see it.

It does seem obvious that when the sound is internal and volume is raised or lowered no voltage adjustment is taking place. Which therefore means the PC is adjusting the soundwaves which I would assume if you keep adjusting gain up and down throughout the signal chain would alter the sound in someway?

Is'nt that why ppl talk about having to set 'unity' gain so that signal in is the same as signal out until it reaches you monitor amp?

I am trying to understand the information provided on bit depth but I do have difficulty taking mathematical information and converting it to solid concepts in my head so bear with me lol

FJ

It really just doesn't matter if you got a 16-bit or above soundcard. The only time bitdepth matters is recording, and even then there are a few techniques to get rid of the noise (if any).

When I said relative I meant that in terms of signal to noise ratio. You can monitor a sound with 120dB of dynamic range at 0dB and at -10dB but it doesn't change the fact that it still has 120dB of dynamic range provided you keep the bit depth at 24 (120dB is the theoretical SNR at this bitdepth but lets just assume it is in practice to keep things tidy). It does not turn to 110dB when you turn down the master gain by 10dB.

quote:

Originally posted by camsr Listen, there's only TWO ways the soundcard is gonna lower the volume: Digitally controlling a solid-state amp circuit on the soundcards output circuit, AFTER the D/A conversion. It would be controlled by software drivers interfacing with the card.

This is exactly the way it works in practice, and is precisely the reason why you do not "lose" resolution no matter what your volume is at.

Go ahead, lower your master volume to 1%, crank your amp to compensate, and play with the fader in Winamp. I'll be shocked if you hear any less of a dynamic range or any sort of distortion, even though technically, by your earlier explanation, there'd only be a few measly bits left at 1%.

But that obviously isn't the case, you've still got all 16 bits, it's just an op-amp after the DAC that's controlling the master volume.

And obviously, even if it worked the other way (digitally attenuating the signal before the DAC), it would still only ever apply to the final output stage, i.e. the transition back to analog domain. This principle has no effect when carrying signals through the digital domain.

So I'll just repeat it again, for anybody who is confused: You do not "lose" bits or resolution when adjusting volume in the digital domain. Ever.

___________________
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

quote:

Originally posted by camsr It really just doesn't matter if you got a 16-bit or above soundcard. The only time bitdepth matters is recording, and even then there are a few techniques to get rid of the noise (if any).

You can never get rid of quantisation noise. It is an inherant part of quantisation. You can however sample so many times per second that 99% of the time you won't be able to tell the difference between an infinitely continuous recording and the same recording after it has been quantised.

If you must truncate bits (for instance going down from a 24 bit recording to a 16 bit recording for an audio CD release) you can 'mask' quantisation noise by dithering which is essentially the process of adding a non cyclical noise source to a recording so that it just about exceeds the quantisation noise ceiling. In this case you will always have low level noise but with dither you can replace that low level noise with a less annoying low level noise.

The human ear tends to be very good at picking up on obviously repeating patterns and we tend to find this annoying or unnatural. Human pattern recognition is in animal terms really very good. Quantisation noise can be annoying because it is cyclical noise that is correlated to the signal. If you bitcrush a sound massively you will notice that the distortion that occurs is extremely uniform. I find massively bitcrushed sounds are incredibly annoying to my ears and the effect is best used sparingly.

Even in a 24 bit recording you can notice quantisation noise - in a complete mix theres often so many things going on that you will never be able to tell but it is noticeable on breaks and fade outs. If you dither sine wave test tones you can also hear quantisation noise and see the noise itself in a spectrum analyser (provided it can monitor sound at that low an amplitude) but I should warn you - you need to amp up your speakers a crazy amount to hear it as the quantisation noise ceiling manifests around -110dB. To listen to it an audible level, a windows system sound could blow your ears off. So be careful if you want to try it out yourself.

Just to be even more pedantic, truncating really means "no dither", just shave off the LSBs. Dithering is an alternative to truncation for reducing the bitrate.

I know, it's semantics, and that's what you meant. Just being an asshole.

___________________
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

You would only use dither when you truncate bits so it doesn't matter. Dithering is not an alternate to truncating bits. It is a process you employ when you have to truncate bits (i.e. rendering down from 32bit float to 16 bit so that it is playable as CD audio) and the reason you would employ it is to add non cyclical noise to the recording to mask the quantisation error that occurs when you suddenly throw away 16 bits worth of dynamic resolution.

But now we are getting off topic so its probably best to just leave it at that. For anyone interested, Izotope wrote a pretty good, no bullshit guide on what dither is and why you would use it. Recommended reading as its actually written in plain english unlike some resources on the subject.

Theres also some info about bitdepth and dynamic resolution which will help those who are still confused about gain control and why bitdepth does not reduce when you lower gain.

You can find the guide on their website and its in PDF format so you need Adobe Acrobat Reader.

Let me point out that there are two different digital domains to be considered here: integer and float. ...and there are two ways to degrade sound quality by digitally amplifying.

You will not loose quality by amplifiying by powers of two if you are in float domain, as this will leave the bits of the mantissa untouched and just increase/decreaese the exponent. In float domain the full scale value is defined to be 1.0 but there is no rule that would prevent your soundcard to produce meaningful voltages above greater values.

In integer domain the maximum integer value represents the maximum voltage your soundcard can output. Any (theoretical) value larger than your available bits will be clipped. Digitally attenuating before the DAC means using less bits no matter what. You WILL loose bits in integer domain as they will be shaved off. This might not be of any relevance if the original signal was normalized and you are attenuating no more than 10dB. But if the original signal was recorded at say -12dB and you digitally attenuate the output by another 12dB (reducing the percieved volume to little less than a half) your SNR will be at best 72dB. Given an average soundcard there will definately be audible noise if analog output is set to produce moderate 95dB(A).

In any digital domain amplifying with factors other than 2 will cause roundoff errors depending on your signal and the resolution of your amplification factor. This is a very slow process but will get get audible eventually... btw: this is one of the reasons some digital filters sound shit.

Production hosts normally operate in 24 or even 32bit floating point domain. So amplification and attenuation might not matter a lot but as soon as you output or render all this will be converted to integer domain.

btw. it is hard to find analog equipment that has a realistic SNR of more than about 95dB. Built in soundcards especially in notebooks are even worse as they will catch digital noise coming from the electronic components of the computer. In PA you migt even find 85dB and less. But Average signal dynamics in electronic music hardly exceed 60dB so you might even get away with 12bits of resolution without hearing quantisation 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 Derivative You would only use dither when you truncate bits so it doesn't matter. Dithering is not an alternate to truncating bits. It is a process you employ when you have to truncate bits (i.e. rendering down from 32bit float to 16 bit so that it is playable as CD audio) and the reason you would employ it is to add non cyclical noise to the recording to mask the quantisation error that occurs when you suddenly throw away 16 bits worth of dynamic resolution.

You're 100% right on concept, just not using the accepted definition of truncate. Truncation literally means you just drop the extra bits. Look it up. In math, truncating a decimal means you just keep the integer part and throw away the decimal, i.e. take the floor even if the value is closer to the ceiling. In audio, it means you just pretend that all the extra bits are zero and slice 'em off.

I'm just nit picking over the definition, that's all. I never said you were wrong.

-
Halo: good point. In 16-bit or 24-bit you're dealing with integers, and 32-bit is floating-point, that's why most sequencers will explicitly say "32 bit (floating point)" in the list of bit rates.

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.

You're also wrong about roundoff error. Take the number "4", which in the digital domain is 100 or 0x4. You can amplify that by a factor of 3, to get 12, which is perfectly represented as 1100b or 0xC. You haven't introduced any error whatsoever, and it goes without saying that you can also divide by 3 without error if your original number was 12.

In fact, integer multiplication in the digital domain is guaranteed to never introduce any error unless you overflow (i.e. go above 0 dB). 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.

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.

___________________
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