recording at -12 db and normalizing to -0.5 db: how much quality is lost?

[trancenonZENsedance]

Hey there!

 

I bought an Allen & Heath Xone:DB2 dj-mixer yesterday and my biggest gripe with it is that when the meters (individual channel as well as master) display +/-0 db, the signal I get when recording in Audacity is actuall at about -12 db. Going by its spec sheet the DB2 works at 48 kHz and 48 bit internally and I assume it receives the 44,1kHz/16bit digital signal from the CDj-400s, upsamples it and then lowers the amplitude by 12 db so there should be practically no quality lost there.

 

I'm just wondering: if I record with the DB2 selected as the source in Audacity, does the software actually receive the 48kHz/48bit signal and convert it to 48kHz/32bit (it is not possible to chooser a higher bit rate than 32 in Audacity) or does it receive "only" an 48kHz/24bit signal (specs say it has 24 bit I/O, but I'm not sure if this applies only to the digital I/Os on the back of the mixer or if it also applies to the digital signal the software receives from the DB2 as a soundcard, instead of the 48 bit EQ)? And also: Do you think Audacitys normalzing algorithm is up to the task?

 

I'm also sending this to the Allen & Heath partner in Austria, so if nobody here has any idea about this, I can still hope to get some infos from them.

[aliendna99]

My first initial reaction is.. 48bit?

CD audio is 16 bit. And I am assuming the music you are working with are normal cd masters? (or equivalent)

[aliendna99]

Ok i reread it twice.

 

Not sure what the question is. Are you asking if audacity lowers the quality? Then no, it doesn't

[frozen dream]

normalizing with audacity is just fine, just listen for bad audio! usually you can hear of something doesn't work out normalized

try recording louder to see of you can crank up the incoming level

[trancenonZENsedance]

Ok i reread it twice.

 

Not sure what the question is. Are you asking if audacity lowers the quality? Then no, it doesn't

 

Yes, that's more or less question 2: I'm not sure if it's theoretically possible to increase the amplitude of a digital audio signal perfectly. If it actually is (which I kind of doubt) then that still doesn't mean the algorithm to do so is perfectly implemented in Audacity.

 

Question 1 would be: Does Audacity receive an 48 bit digital audio stream to work with from the DB2 or a 24 bit stream?

 

normalizing with audacity is just fine, just listen for bad audio! usually you can hear of something doesn't work out normalized

try recording louder to see of you can crank up the incoming level

 

The problem with recording louder is that it becomes harder to match the levels of the tracks by looking at the meters of the DB2, since the master vu-meter on it goes [...], -3, -2, -1, 0, +1, +2, +3, +6 db and then there's the top led to indicate clipping. If I record at a louder volume, then I cannot match levels as closely, which leads to greater variation, which leads to a generally lower level after normalizing.

 

Initially I was pretty stressed out about the DB2 delivering -12 db (after all the only reason I got a digital mixer in the first place was to achieve optimal sound quality) and I personally would still prefer around -3 db, but I know how many DJs don't mind playing deeply into the red part of the meters at all, so I guess I can't really blame A&H for not letting some folks completely f' up the sound quality that easily, when the trade off is relatively small...

[aliendna99]

It is perfectly safe to normalize. I don't even think any algorithm is used that is unique to the different waw editors.

It just puts the max peak of the wav to whatever db you choose thus increasing volume.

Normalizing the rms (or loudness) is another story.

 

 

 

Audacity will receive a 24bit signal. It doesn't convert it from 48. It just cuts or removes the data from 24 to 48. (kinda the same as writing 0 in that data range)

 

You will most likely not lose any quality doing this since the audio data is 16 bit to begin with making it very unlikely that there is any data in the 24-48 bit range anyway.

[trancenonZENsedance]

Thanks for the answers! I don't even know what the difference between normalizing the max peak of a wav and normalizing rms is, mathematically. I guess it's best to set Audacity to record at 48 kHz and 24 bit then, rather than 32 bit.

[aliendna99]

There is a huge difference between normalizing peak vs. normalizing rms.

 

 

You could use 32bit also if you want but personally I wouldn't. Seems unnecessary.

If you want absolute quality you must convert the bitrate from 48 to 24 Before it enters auadicity.

For that you will need an extra peice of equipment.

Couldn't you set your equipment to work in 24bits instead of 48 btw?

 

Either way it's all just for peace of mind the end result will be the same to something like 99.99999%

And it will definitely not be any audible difference in the end result at 44 kHz / 16 bit. (cd master)

[trancenonZENsedance]

I don't think it's possible to set the xone:db2 to work at 24 bit instead of 48 bit internally.

 

Thanks a lot for your input, aliendna99!

[Dolmot]

What do you use to connect the CDs to the mixer and the mixer to the computer?

[trancenonZENsedance]

The CDj-400s are connected to the mixer (digital out on the cdjs -> digital in on the mixer) by 75ohm coaxial cinch cables, the mixer is connected to the PC per USB.

[Dolmot]

One spec page I found says "Built into the Xone:DB2 is a fantastic-sounding 24-bit/48kHz USB 2.0 audio interface, which lets you send and receive up to four stereo channels to and from your computer" which would imply that it's 24-bit fixed point you're receiving. If you use that as the recording format, in theory it's an exact match to the data transmitted via USB. 32-bit float should be just fine too.

 

If your data is at -12 dB peak, to my best knowledge it means that approximately the highest two bits are empty but you still have 22 bits of information. In other words, if you ultimately distribute in a 16-bit format, you still shave about six bits (36 dB) of information in the final output stage. The content of those is debatable if the input was a mixture of 16-bit CDs. Of course, if one CD was playing at another -12 dB compared to the overall level, its lowest bits would produce genuine information to the lower bits of the 24-bit recording. Effects complicate the matter further. On the other hand, it's eventually lost in the output, and there's very little proof of anyone being able to hear dynamics beyond 16 bits anyway.

 

If it all goes as intended, you have plenty of extra for processing, and the digital chain giving you potentially 22 bits is probably far better than most home mix recordings ever produced. It exceeds the source, the target, and human hearing. No reason to worry about the two-bit headroom. I'd say that at that point the real error is introduced by sample rate conversions (pitch changes, 44.1 -> 48, possibly back again) which cannot be done 100% perfectly by any means.

 

Peak normalisation in Audacity can be done with "Amplify" too. See http://wiki.audacityteam.org/wiki/Amplify_and_Normalize for some documentation.

[aliendna99]

It being 24bit rather than 48bit (which I never even heard of before) makes a lot more sense.

 

Regarding the - 12db "issue" it is most likely just as dolmot describes it.

This is not bad at all, but rather quite good as it leaves a lot of dynamic headroom for mastering.

As i understand it your "issue" is that the highest peak is at - 12db correct?

 

 

(

My mixdowns are usually around the - 15db (rms) mark to give enough room for mastering where the final product is something like -8db rms.

Max peak is something like -2db before and - 0.01db when finished.

)

 

 

 

It could also just be something you have overlooked in your settings.

[trancenonZENsedance]

Dolmot: What you write makes sense to me. I don't understand this part though:

 

Of course, if one CD was playing at another -12 dB compared to the overall level, its lowest bits would produce genuine information to the lower bits of the 24-bit recording. Effects complicate the matter further.

 

Peak normalisation in Audacity can be done with "Amplify" too. See http://wiki.audacityteam.org/wiki/Amplify_and_Normalize for some documentation.

 

When normalizing in Audacity, would you recommend enabling DC offset correction? I'm assuming it's not possible to introduce DC offset when staying inside the digital realm all the way from the source to the recorded .wav and I'd assume most commercially released tracks would have any DC offset removed already; but could it theoretically hurt to do it a second time (as in: could there be something that looks like DC offset, but is actually intended and/or something different than DC offset)?

 

 

As i understand it your "issue" is that the highest peak is at - 12db correct?

 

Exactly this. I have never been mastering my mixes. I don't even really know what mastering means in relation to a dj-mix (as opposed to a studio-mixed album or compilation). When recording an analogue signal, I tend to try to get as close to 0 db as possible during the loudest part of a track, while avoiding signal clipping to happen more often than maybe once or twice per minute (assuming one occurance of clipping is one vertical red line at default zoom in Audacity).

It could also just be something you have overlooked in your settings.

 

I think I looked through everything which had any chance to be even remotely relevant. Probably twice.

[aliendna99]

Then in your case I would just normalize peak to - 0.01 and smile and be happy and think nothing more of it.

Could you do me a favor though, after you normalize, could you please analyze the wav for the average rms value.

 

I am a little worried that either your equipment or auadicity thinks that - 12db actually is -0db.

If this is the case there is not a real dynamic and the issue is real and should be fixed.

I'm not sure you understand what I mean it's difficult to explain.

Could you also take a screenshot of the raw wave before you touched it? It's quiet easy to spot if it looks okay

[Veracohr]

Welcome to the wonderful world of decibels!

 

First thing to remember is that dB is a relative term; that is, X dB is always measured relative to some reference point. On page 11 of the manual for your mixer, it says "The meter reads '0' for an XLR output of +4dBu." +4dBu is a common output level reference in audio equipment, and 0dBu is defined as SQRT(0.6)V RMS, or about 0.7746V RMS. So dBu is an electrical reference.

 

Now, the meter in your DAW is showing dBFS, or decibels-full scale. "Full scale" in this case means the maximum amplitude in a digital signal, when all bits are 1. So -12dB in your DAW means it is 12dB below maximum amplitude. -12dB is a good level for audio to be coming into your DAW. Also, on page 37 of the mixer's manual where it says "Digital Architecture Specification", it states "Analogue/Digital Line-up: +12dBu=0dBFS". Don't worry, your equipment is all fine and working according to spec.

[trancenonZENsedance]

I recorded Leftism - Mind Doctor (http://www.ektoplazm.com/free-music/labyrinth-of-your-mind).

 

Screenshot of Audacity before normalizing:

 

 

 

Aufter normalizing to -0.01 dB the result of contrast analysis for the upper (left?) channel of the entire track is -13,8 dB.

[trancenonZENsedance]

[...] Don't worry, your equipment is all fine and working according to spec.

 

So essentially this means recording at -12 dB and then normalizing the result really close to maximum amplitude is more or less the optimal thing to do. And as I understand it from your post, this is the case because if there was the "loudest theoretically possible sound" on a cd somewhere it would put the signal to the maximal amplitude, but not yet into clipping?

[Veracohr]

As far as what's optimal, that depends on your goal. What are you doing with the audio once it's in Audacity?

[trancenonZENsedance]

Mostly I just archive it to the hard disk as a 44.1kHz/16bit wav-file. Sometimes I copy it onto a portable music player, burn an audio cd or put it on a usb stick. I'm planning to upload some mixes to soundcloud or a similar service, as soon as I have some that I feel are worthy of that. 48kHz/24bit might sound slightly better but 99.5% of people wouldn't care about that (assuming there is a service that would host such files) and even I personally haven't bothered with higher-resolution-than-redbook formats so far.

[trancenonZENsedance]

And as I understand it from your post, this is the case because if there was the "loudest theoretically possible sound" on a cd somewhere it would put the signal to the maximal amplitude, but not yet into clipping?

 

Uhm, something seems to be off in my understanding: Isn't most of the goa/psy stuff that gets released nowadays already normalized to be as close to 0dB as possible? If that's the case then the loudest theoretically possible signal would be hardly louder than what's already on the cd - and I still only get up to -12 dB.

[aliendna99]

I am not sure I am understanding the screenshot . I've never used auadicity but it looks strange to me.

Could you try again maybe a bit more zoomed in?

Also the wav seems centered around zero where 1.0 is the ceiling.

That is also Wierd because the ceiling should always be 0 as anything above that is clipping.

So I'm not sure what the numbers there mean.

 

 

And could you check the rms after normalizing.

[aliendna99]

And yes all cds are touching almost the 0db

[trancenonZENsedance]

I don't manage to get Audacity to record something right now (this has happened once before, I tried some things, then it worked again, but I didn't really gather what actually made it work again). I'm working on that right now.

 

Aufter normalizing to -0.01 dB the result of contrast analysis for the upper (left?) channel of the entire track is -13,8 dB.

 

I think contrast analysis is what returns average RMS in Audacity.

[Dolmot]

Dolmot: What you write makes sense to me. I don't understand this part though:

 

Well, I think the point was something like this. Let's start with the basics. The following three operations are supposed to be equivalent:

- Doubling the magnitude (sample values)

- Multiplying the signal energy by four

- Increasing its power by 6 dB (OK, 10*log10(4) = 6.0206... but we can call it six.)

 

And the same in reverse (half magnitude, quarter energy, -6 dB).

 

Let's simplify the matter and assume that you're recording a 16-bit CD to a 24-bit wave 100% digitally in the same sample rate and at quarter magnitude so that its peaks are at 0.25 level (-12 dB). For further simplification, we'll assume that signals are stored as a sign bit and N-1 magnitude bits.

 

The bit pattern would look like this, with the sign and the most significant bits on the left.

 

S..XXXXX XXXXXXX XX......

 

After the sign bit, the leftmost two bits are hard zeros. Then you have 15 bits of magnitude information. The rightmost six could be empty, but in your case they would probably contain something from sample rate conversion and other DSP. Not any real low-energy information of the music, though, because that didn't exist in the source.

 

But if you mix in another CD at 1/4 magnitude compared to the first one, its bits would look like this in the 24-bit recording

 

S....YYY YYYYYYYY YYYY....

 

for the same reasons. Because the recorded signal is the sum of both, you'd have genuine audio content in sign and bits 4-20 (so 18 bits).

 

S..XXZZZ ZZZZZZZZ ZZYY....

 

However, the bits 19-20 of the first signal are effectively quantisation noise or otherwise meaningless so they'd largely mask anything happening in the second signal. Besides, bits 4-18 and sign already contain ~96 dB of dynamic range which is enough for human hearing, and you're supposed to keep only those for a 16-bit output. The lowest bits of the second signal are masked, meaningless, and eventually discarded.

 

This is all largely theoretical. As stated earlier, you have interpolation noise from pitching and sample rate conversions so what you get is no longer even fully accurate 16-bit information. And CDs today go through so much compression, soft-clipping and reshaping that there isn't 96 dB of actual dynamics to begin with. It can be awfully distorted by excessive processing and loudness war. You won't really get any improvement by saving the final output at over 16 bits in this kind of scenario. For intermediate work copies it may be a good idea to save at 24, though.

 

The end result stands. There's sufficient extra at both ends. Don't worry about things happening below the first (meaningful) 16 bits.

 

If you want perfect replication with absolutely no clipping or signal modification, normalise the peak to max sample value. Sometimes it's a good idea to shave a few odd peaks to get the "actual" signal to the full range of sample values. Some prefer even more compression or levelling to compensate incidental level variation between tracks. DC shouldn't exist in digital recordings. Use DC removal only if you have a reason to suspect it being present.

 

It's just a mix tape. While I store my bought music in CD-quality lossless, I can happily listen to mixes recorded from worn cassettes or 112 kbps 1st generation mp3s if the content is right...