Does Traktor's master gain have any effect on recording gain?

Home :: General Discussion :: Does Traktor's master gain have any effect on recording gain?Reply
Does Traktor's master gain have any effect on recording gain?
Posted on: 12.06.2012 by Madelene Witek
I know you should record at ~-10dB, does it matter what the master gain is if i'm recording internally and have the recording gain set -10dB? I had both at -10 but didn't know if that would make the entire thing REALLY quiet.
Dorie Scelzo
14.06.2012
Originally Posted by rgtb
traktor's internal representation is irrelevant in this context. the relevant signal which the audio recorder records is a quantized 16-bit integer signal. if you set the gain in the audio recorder such that your highest peak is -18dBFS, then you're throwing away 3-bit of resolution. the three most significant bits of each sample will be zero, *always*.
That's not entirely correct. Traktor's internal representation is relevant because it allows you to use completely different levels for recording and output, which was part of the point of this thread.

One clarification…you're not losing the most significant bits…you're losing the least significant bits (the ones that get cut off below the noise floor of digital audio). When you then master or normalize the recording, you wind up turning those bottom 3 bits into noise either from your mastering processors or from the normalization algorithm…both of which probably don't actually add noise…but let's pretend they do and that the noise is completely uncorrelated to the audio. That means that the loudest possible signal of the noise in those 3 bits is 0x0008 out of 0xFFFF (a ratio of .000107).

Now, when we're actually playing the audio, it has to go through a DAC and get converted to an analog voltage signal. The math gets weird here, and it's entirely possible that I'm doing something wrong. If there's an electrical engineer out there that can correct me, please do. The actual numbers (sample values) involved in 16-bit digital audio (0x0000 to 0xFFFF) are linearly related to output voltage, which is (eventually) linearly related to speaker movement and sound pressure. Your ears hear logorithmically, and that's how the dBFS scale works too.

In a Studio:

AES standard calibration gives 0dBFS = +24dBu = 1.22 V. (comes from +4 dBu = -20 dBFS reference level for studio gear )
Your noise (representing .000107 of Full Scale) is at .000107 * 1.22 = 0.00013 V = 0.13 mV = -75.5 dBu.
So, the noise is actually about 100dB quieter than your peak.

That's really quiet. And, it actually makes sense if you take that article as true and the dynamic range of 16-bit audio actually is more like 120dB than 96dB. I love it when things are internally consistent.

So, is it ideal? Probably not. But it's about as loud as someone trying to talk to you in a normal voice while operating a chainsaw without earplugs.

Originally Posted by rgtb
however, i do conjecture scenarios exist where you can ABX between a recording made at close to 0dBFS and a recording at -40dBFS which we later bring up to the same volume as the first recording. in the end, it's an empirical question. you need to do ABX tests.
If you're recording in 16-bit…yes. But no one would do that unless they didn't have any other choice…which you never do. All it's telling me is never to bother with Traktor's internal recording. But still…16-bit audio is plenty wide for this type of recording.

Originally Posted by rgtb
today's productions are compressed a lot and they are clipped a lot. we're in the late stage of the loudness war. the stuff mastering engineers do will bring up productions' noise floor. once again, i'm not sure what that means for us as DJs.
For recording…it's simple.

Don't clip.

That's basically it. You don't want your RMS level to be -50 dBFS. So don't be crazy. But you don't need to jump through hoops and make sure you know where your peaks are to make sure that you're getting a recording as loud as you can without clipping. Just turn it down enough to make sure you don't clip and don't worry about it. Peaking at -5dB is probably fine.
Dorie Scelzo
13.06.2012
Originally Posted by rgtb
that's half of the story. the other half of the story is that peak levels in the audio recorder should reach close to 0dBFS. i mean, if your peaks are at -3dBFS, it's probably nothing to worry about. but if your peaks are at -20 or -25dBFS, you're wasting bit resolution.
That's really not that big of a concern. You don't want to record at -70dB or anything crazy because there's no reason, but there are 3 big players in this.

  1. Modern music is really compressed. You only need a few bits worth of dynamic range to represent everything except fades to silence. That matters if you're recording instruments…but it's not as big of a deal for full tracks.
  2. The general rule of thumb is that 6dBFS = 1 bit (for integer representations…floating point numbers work differently). But that's based on full spectrum audio (white noise). 1 bit can represent a good bit more dynamic range than that for anything that isn't white noise…which depends a lot on the frequency content of the music.
  3. Traktor uses 32-bit flaot representations internally…which have a noise floor that's basically inaudible (around -1500dB).


What that means is that you can run your Master wherever the hell you like it (goal: don't clip) and amp your final output after the DAC to get it to a reasonable level. I did a set on Traktor at -40dBFS just for the heck of it. The difference wasn't audible.

Your recording level should be reasonable, but if you're recording a 24-bit file…peaking at -10dB or -20dB is much better than aiming at -3dB and clipping when you boost an EQ knob a bit too much. Even at 16-bit, it's fine. Read this and pay attention to the section on dynamic range a couple pages down.

Because of the way digital audio and the human ear works, 16-bit audio can record things a lot quieter than -96dB. He makes a legit claim that -120dB is a much closer figure, and as he shows, "120dB is greater than the difference between a mosquito somewhere in the same room and a jackhammer a foot away.... or the difference between a deserted 'soundproof' room and a sound loud enough to cause hearing damage in seconds."

That's what you're working with.

10 or 20dB at the top doesn't matter, and my experiments have borne this out in practice.

"As loud as you can without clipping" died with tape. Especially if you're recording in the box and not going through converters, the only thing that really matters is "don't clip." If you can see it on the meter, you can record it. Yeah…when you master/normalize it, you'll be amplifying the noise floor…but Traktor's noise floor is inaudible anyway. As long as you're using something modern for your mastering/normalizing, everything should be fine.

One recent interesting realization from these experiments: Maschine's panning doesn't go to -infinity. If you pan a sound hard left…the right side still carries the full signal. If you host it in a DAW set to use 32-bit float numbers…neither one has a nose floor to speak of. Adding around 150dB of gain to the right side of a stereo signal panned hard left that should be silent gives you the right side of the original…in what seems to be perfect fidelity.

That basically means that Maschine can output at around -150dBFS and as long as you're getting it's 32-bit float internal representation out of it and your gain stage is perfectly clean (which DAW gain utilities are)…the signal seems like it's still perfect.

It really makes me wonder why Trakor defaults to playing everything so loud. The only thing that really matters is not clipping your sound card outputs and having a quiet enough gain stage after the DAC to get it up to a reasonable level before it hits the amplifiers.
Madelene Witek
12.06.2012		I know you should record at ~-10dB, does it matter what the master gain is if i'm recording internally and have the recording gain set -10dB? I had both at -10 but didn't know if that would make the entire thing REALLY quiet.
Dorie Scelzo
14.06.2012		That's not quantization noise. Quantization errors come out of how bit depth and sampling frequency interact. You "fight" it by adding virtually inaudible noise (dithering)., which traktor doesn't do.

-50 is absurd, but I still believe keeping your levels around -18 is a better choice than right around 0.
Dorie Scelzo
14.06.2012
Originally Posted by rgtb
traktor's internal representation is irrelevant in this context. the relevant signal which the audio recorder records is a quantized 16-bit integer signal. if you set the gain in the audio recorder such that your highest peak is -18dBFS, then you're throwing away 3-bit of resolution. the three most significant bits of each sample will be zero, *always*.
That's not entirely correct. Traktor's internal representation is relevant because it allows you to use completely different levels for recording and output, which was part of the point of this thread.

One clarification…you're not losing the most significant bits…you're losing the least significant bits (the ones that get cut off below the noise floor of digital audio). When you then master or normalize the recording, you wind up turning those bottom 3 bits into noise either from your mastering processors or from the normalization algorithm…both of which probably don't actually add noise…but let's pretend they do and that the noise is completely uncorrelated to the audio. That means that the loudest possible signal of the noise in those 3 bits is 0x0008 out of 0xFFFF (a ratio of .000107).

Now, when we're actually playing the audio, it has to go through a DAC and get converted to an analog voltage signal. The math gets weird here, and it's entirely possible that I'm doing something wrong. If there's an electrical engineer out there that can correct me, please do. The actual numbers (sample values) involved in 16-bit digital audio (0x0000 to 0xFFFF) are linearly related to output voltage, which is (eventually) linearly related to speaker movement and sound pressure. Your ears hear logorithmically, and that's how the dBFS scale works too.

In a Studio:

AES standard calibration gives 0dBFS = +24dBu = 1.22 V. (comes from +4 dBu = -20 dBFS reference level for studio gear )
Your noise (representing .000107 of Full Scale) is at .000107 * 1.22 = 0.00013 V = 0.13 mV = -75.5 dBu.
So, the noise is actually about 100dB quieter than your peak.

That's really quiet. And, it actually makes sense if you take that article as true and the dynamic range of 16-bit audio actually is more like 120dB than 96dB. I love it when things are internally consistent.

So, is it ideal? Probably not. But it's about as loud as someone trying to talk to you in a normal voice while operating a chainsaw without earplugs.

Originally Posted by rgtb
however, i do conjecture scenarios exist where you can ABX between a recording made at close to 0dBFS and a recording at -40dBFS which we later bring up to the same volume as the first recording. in the end, it's an empirical question. you need to do ABX tests.
If you're recording in 16-bit…yes. But no one would do that unless they didn't have any other choice…which you never do. All it's telling me is never to bother with Traktor's internal recording. But still…16-bit audio is plenty wide for this type of recording.

Originally Posted by rgtb
today's productions are compressed a lot and they are clipped a lot. we're in the late stage of the loudness war. the stuff mastering engineers do will bring up productions' noise floor. once again, i'm not sure what that means for us as DJs.
For recording…it's simple.

Don't clip.

That's basically it. You don't want your RMS level to be -50 dBFS. So don't be crazy. But you don't need to jump through hoops and make sure you know where your peaks are to make sure that you're getting a recording as loud as you can without clipping. Just turn it down enough to make sure you don't clip and don't worry about it. Peaking at -5dB is probably fine.
Dorie Scelzo
13.06.2012
Originally Posted by rgtb
that's half of the story. the other half of the story is that peak levels in the audio recorder should reach close to 0dBFS. i mean, if your peaks are at -3dBFS, it's probably nothing to worry about. but if your peaks are at -20 or -25dBFS, you're wasting bit resolution.
That's really not that big of a concern. You don't want to record at -70dB or anything crazy because there's no reason, but there are 3 big players in this.

  1. Modern music is really compressed. You only need a few bits worth of dynamic range to represent everything except fades to silence. That matters if you're recording instruments…but it's not as big of a deal for full tracks.
  2. The general rule of thumb is that 6dBFS = 1 bit (for integer representations…floating point numbers work differently). But that's based on full spectrum audio (white noise). 1 bit can represent a good bit more dynamic range than that for anything that isn't white noise…which depends a lot on the frequency content of the music.
  3. Traktor uses 32-bit flaot representations internally…which have a noise floor that's basically inaudible (around -1500dB).


What that means is that you can run your Master wherever the hell you like it (goal: don't clip) and amp your final output after the DAC to get it to a reasonable level. I did a set on Traktor at -40dBFS just for the heck of it. The difference wasn't audible.

Your recording level should be reasonable, but if you're recording a 24-bit file…peaking at -10dB or -20dB is much better than aiming at -3dB and clipping when you boost an EQ knob a bit too much. Even at 16-bit, it's fine. Read this and pay attention to the section on dynamic range a couple pages down.

Because of the way digital audio and the human ear works, 16-bit audio can record things a lot quieter than -96dB. He makes a legit claim that -120dB is a much closer figure, and as he shows, "120dB is greater than the difference between a mosquito somewhere in the same room and a jackhammer a foot away.... or the difference between a deserted 'soundproof' room and a sound loud enough to cause hearing damage in seconds."

That's what you're working with.

10 or 20dB at the top doesn't matter, and my experiments have borne this out in practice.

"As loud as you can without clipping" died with tape. Especially if you're recording in the box and not going through converters, the only thing that really matters is "don't clip." If you can see it on the meter, you can record it. Yeah…when you master/normalize it, you'll be amplifying the noise floor…but Traktor's noise floor is inaudible anyway. As long as you're using something modern for your mastering/normalizing, everything should be fine.

One recent interesting realization from these experiments: Maschine's panning doesn't go to -infinity. If you pan a sound hard left…the right side still carries the full signal. If you host it in a DAW set to use 32-bit float numbers…neither one has a nose floor to speak of. Adding around 150dB of gain to the right side of a stereo signal panned hard left that should be silent gives you the right side of the original…in what seems to be perfect fidelity.

That basically means that Maschine can output at around -150dBFS and as long as you're getting it's 32-bit float internal representation out of it and your gain stage is perfectly clean (which DAW gain utilities are)…the signal seems like it's still perfect.

It really makes me wonder why Trakor defaults to playing everything so loud. The only thing that really matters is not clipping your sound card outputs and having a quiet enough gain stage after the DAC to get it up to a reasonable level before it hits the amplifiers.
Latoria Kavulich
13.06.2012		just keep both levels out of the red.
Nedra Fresneda
12.06.2012		Nope, the internal recorder has it's own gain knob.
Deeann Cheron
12.06.2012		where did you get -10db from?

I have my mix recorder (internal) set at -3 and my master set at 12 o'clock

<< Back to General DiscussionReply

Copyright 2012-2023
DJRANKINGS.ORG n.g.o.
Chuo-ku, Osaka, Japan

Created by Ajaxel CMS

Terms & Privacy