The noisy truth about your DAW: An ear-opening investigation
Comments on this video
You can comment on this video at YouTube
You can comment on this video at YouTube
Thursday April 20, 2023
David Mellor
David Mellor is CEO and Course Director of Audio Masterclass. David has designed courses in audio education and training since 1986 and is the publisher and principal writer of Adventures In Audio.
@bahathir_: Hi,
If you still have a tape recorder, record the recoder's noise floor and lets other apriciate the technology we currently have. :)
Thankyou.
@nate_d376: #BringBackMono ! Lol
@j7ndominica051: The DAW, a plugin or both may inject a floating-point "denormal" noise between -150 and -200 db, usually a square wave bias, which is designed to prevent the signal from becoming too low amplitude and costing more operations for the CPU. There is a setting for it in Reaper. I think it's not quite the same as a plugin not having a footroom because of flaws that accumulate throughout its working stages.
@imqqmi replies to @j7ndominica051: Yeah 32 floats have about 6 positions and an exponent component. It means on very loud sound you could loose very quiet squiggles omposed over it. I mean 123456^33 + 1, the plus 1 is lost. Same with 0.000000123456+1, 123456 is lost. That's why the daw limits the values to 150 dbfs, about 33 milion steps. I bet if youdo a 200hz bass sinewave at 150db amplitude and add a sinewave at 10k at 144db lower amplitude, that 10k signal is lost.
@paulstubbs7678: So what are the limit's?
is this the 32bit float format that one hears of
@AudioMasterclass replies to @paulstubbs7678: There's probably no limit other than the laws of physics. 32-bit float is great for production but not for the end product where 24-bit is more appropriate. 32-bit float has no better resolution. I think that 24-bit audio should be good enough for anyone, no matter how golden their ears. Having said that, I'd like to see a transition to lossless streaming, and a sampling rate of 96 kHz. DM
@nicoras8803: From a theoretical engineering point of view, this was actually quite bizarre and interesting. Spent 50 years in analogue signal processing, noise was our greatest enemy. I take it that turning down the gain compressed the noise floor inherent in the signal by the same amount, and we are not considering the actual noise floor or sensitivity of the equipment. I speculate that not having a constant signal i.e. 220Hz does have something to do with the fact that noise does not correlate incoherently while the signal is coherent. Interesting topics you cover - thanks.
@jackevans2386 replies to @nicoras8803: "we are not considering the actual noise floor or sensitivity of the equipment" Totally !
@isotoxin: Isn't it the case that noise is presented by audio interface amplifiers and not by the DAW application?
@AudioMasterclass replies to @isotoxin: The DAW works on maths and there's no noise in maths, not as far as I'm aware anyway. But getting the audio into and out of the DAW will always have noise so it's inherent in the process. DM
@RocknRollkat: Hello DM,
To elaborate on my comment from last month, I did these tests a few months back while proving a point about that -18 dB fallacy.
I recorded a pure 1kHz 0 dB sine wave into Wavelab, reduced to level to -60 dB, and printed the .wav file.
I loaded the -60 dB file back into Wavelab and 'normalized' it up to 0 dBs.
I got the predicted results, plenty of distortion.
So yes, even though your DAW preserves the original levels without telling you (or me), the truth comes out when you bring a truly low level wave up to normal volume and out into the real (analogue) world.
That's the part the youngsters don't understand.
As a side note, Wavelab 6.1 does all internal processing in 24 bit to maintain as good quality as possible.
I suspect most DAW do the same.
Best regards,
Bill P.
@searchiemusic replies to @RocknRollkat: i've literally never had this problem, especially at 60db, that's nowhere near quantization error levels with 16 bit+ audio, i'm guessing you truncated it without dithering it, that's literally the only way you could have gotten distortion
@RocknRollkat replies to @RocknRollkat: @@searchiemusic I created the problem to prove a point. Remember, no matter what you're processing at internally, your end result gets printed to CD at 16 bits, with the inherent -96 dB noise floor.
It's easy to duplicate the experiment, give it a try.
@JeffWernerIthacaNY replies to @RocknRollkat: The noise was introduced in the wav file format you saved, If you had saved the intermediate wav file in 32 bit floating point format you wouldn’t have gotten any noise from that experiment. Almost all modern daws work in 32 bit float nowadays to make this problem totally irrelevant. Very convenient!
@RocknRollkat replies to @RocknRollkat: @@JeffWernerIthacaNY The whole point of my test was to show that noise does exist when you print to CD, which as you know is 16 bit.
So yes, giving up 3 bits really DOES make a difference in the end product.
@JM_2019: Isn‘t it pretty obvious that mathematical operations don‘t introduce noise?
@Bluelagoonstudios: You should consider to change to Reaper, the stock Trim plugin goes from -150 to +150 dbfs. For recorded tracks that have noise in it, or ground loops in a guitar amp, we use a gate. Like the XLA-3 compressor, we can add noise, or a tape plugin. Which are very good these days.
@Tachy_Bunker: 32-bit float has a range of 1,528 dB's. If you were in space, you would hear things louder than 32-bit noise floor 😂
@sonicsaviouryouwillnotgetm6678: what is the reason? Internal 64bit processing, I presume?
@Emlizardo: You had me at Marianas Trench.
@DiegoINSOMNIA: Can you make a video about aliasing, saturation and compression? Problems and tips about it?
@Paul58069: Eric Idle strikes again :)
Jokes aside, excellent series and quite enjoyable presentation !
@Mikexception: Volume controll with digital do not involve any very low digital signals but only specified numbers for very low DA output. DA noise exists anyway but is not being recorded, only is each time created from scratch from digital data. so in digital recording at level -135 nowhere is recorded any -135dB by volume electrical signal - it is recorded with full standard voltage. hat is why I am not surprised by conclusions.
Sensation of "sound of silence" may be experienced also in -50dB analog level Because in my case I always or almost always hear sound layer from neighbours in the same block, from street which i very quiet but birds sing, fly, my computer working and my ears ringing. The level of perceived silence s too loud to hear any noise at -50dB which is generaly considered not great for analog.
@IntheDAW: This was amazing
@herbst1398: Hey, im here to tell you, yes a video about gain staging would be great! - > in my case, a guitar ; )
@Alej_915: A+ for the james cameron reference. Cheeky bastard 😂
@RocknRollkat: The test proves nothing about system noise floors.
Obviously your DAW preserves the original 0 dB signal and recalls it.
Try recording a sine wave from an external signal source at -138 dB, then normalize it to 0 dB and we'll see what's what.
Bill P.
@AlexLapugean replies to @RocknRollkat: But ... that was not the point of the video ... why even bring this up? The video was explicitly and exclusively about the DAW, if you should, of should not worry about dynamic range/reducing/amplifying sounds inside of your DAW. Smartass...
@ferociousmullet9287: 1 bit = 6.06dB so you're reducing it by 23bit = 139.38dB. Not exactly 139dB you are then adding 139.38dB back giving you an extra 0.76dB which is probably being rounded down to 0.6dB by the display. So the discrepancy is valid, the amount of error being shown is slightly low due to I feel a lack of precision in the scaling on the display. Those bars are only 0.2dB in precision, so it's rounding down to 0.6 not up to 0.8 would be my guess.
@bobwatkins1271 replies to @ferociousmullet9287: Exactly. Attenuating the signal as he did introduces quantization error -- anything between 0 and -6dBFS would be lumped into that single least-significant bit. You can reduce the quantization error by dithering, but then you introduce noise.
@AlexLapugean replies to @ferociousmullet9287: How is he reducing the signal by 139.38dB when pulling the fader down to -138dB? I fail to understand what you mean...
@bobwatkins1271 replies to @ferociousmullet9287: @@AlexLapugean He scaling the signal by 2^-23. In decibels, that's 20 Log10(2^-23) = -138.47dB. This is equivalent to shifting each 24-bit sample right by 23 places, preserving only the MSB.
@AlexLapugean replies to @ferociousmullet9287: @@bobwatkins1271 But why do you say he is scaling by that amount, do you say you can only scale by 2 to a whole power? That makes no sense, by that logic, you only have 23 possible steps between 0 and -138dB, I am pretty sure that is not true. Also ... why 24 bit? That is only applicable if the recording is 24bit, and even then, the processing inside of a daw is done at least at 32bit float, or even 64 in some daws.
@bobwatkins1271 replies to @ferociousmullet9287: @@AlexLapugean You can scale by any amount, of course, but -138dB is the example he used, and that's approximately a 23-bit shift. Ultimately, the sound will be exported to media as integer samples.
@dreamscuba: Very interesting topic and experiment. Thanks
@christopherward5065: That was a great demo. I thought the signal would come back with some grunge from the noise floor.
@thorstenoerts: The wonders of floating-point arithmetic.
@melaniezette886 replies to @thorstenoerts: Yes, floating point is a kind of magic strange world...
@bestdisco1979: I seriously don’t get gain staging. Surely you need need enough gain to sufficiently feed your plugin chain , if I try the unity gain thing I don’t seem to have enough and end up with some faders way down low. Where am I going wrong ?
@TheNoiseFloorav replies to @bestdisco1979: I made a video explaining everything you need to know in simple to understand language. - https://youtu.be/49Woh9DleSM
@RedBlueSam replies to @bestdisco1979: Set a trim plugin as first insert on your mixer channels and set the appropriate amount of gain that way instead of doing everything with your faders. The faders affect the signal post the plugin chain.
@TheNoiseFloorav replies to @bestdisco1979: @@RedBlueSam There's zero reason to need a trim plugin as the first insert in your DAW, and the faders have zero impact on the signal after the plugin chain in a DAW.
@RedBlueSam replies to @bestdisco1979: @@TheNoiseFloorav "There's zero reason" if you only record live instruments and set the appropriate amount of gain during the recording stage. If you're working with digital samples in electronic music, many of them are normalized so then it can be useful (not a necessity) to use a trim plugin beforehand. In FL Studio you can also set gain via the channel rack (pre mixer insert/plugin chain) so you won't need a trim plugin in that case (I guess this depends on your daw).
As for your last point, your faders regulate the gain input to the stereo buss, so if you have a plugin chain there, the amount of gain from the different mixer channels will affect the way their signals are processed on the stereo buss.
@TheNoiseFloorav replies to @bestdisco1979: @@RedBlueSam Useful is something that applies to one's workflow preferences. You said yourself that it's not a necessity.
You may prefer working that way. That may make sense to you, and I won't tell you to stop. As someone who isn't neurotypical, I'm not gonna tell you how to do things if that way makes sense to your specific way of thinking.
But what you're saying doesn't have a reason based on science, only preference.
@teashea1: very good ---- video on gainstaging would be great
@Pete731 replies to @teashea1: +1 for gainstaging
@jeffreyhanc1711 replies to @teashea1: +1
@intheblink: I really appreciate your approach! Great work! Analog is great and all and sometimes a little noise is nice, but it can be such a pain to deal with. Thank goodness we have such incredible tools to deal with it these days.
@TheNoiseFloorav: Thank you for publishing accurate info
@thegroove2000: Thanks Paul.