Importance of Recording Levels for 16-Bit?

Special Interest Groups Recording Techniques
1

I’m going to be recording some vinyl LPs using a 16-bit USB audio interface (it does have a knob controlling the level that enters its ADC). Apart from noise concerns, is it true that recording at a higher level will give a more accurate recording than at a lower level? The example below made me think so:

– Will use 16 bit in the example: 16 bit has 2^16 baskets of amplitude = 65536, which covers -1 to 0 to +1 on the Audacity linear scale. For simplicity consider only the positive part of the signal, which is covered by half that number, so 32768 baskets of amplitude.
– The example: an analog signal with a positive peak at +0.3 on the linear scale (assume negative peak -0.3), being converted to digital with 16 bits would be documented by 0.332768 = 9830 baskets of amplitude. However if that analog signal were first amplified to peak at +0.9, then it would be documented by 0.932768 = 29491 baskets of amplitude. So it would seem that tripling the analog signal before entering the ADC would document the same wave with three times the baskets of amplitude, or three times the fineness.

I initially thought 9830 baskets still seemed like a large number, but then remembered reading that some of the very first CDs were digitized with 14 bits and often sounded bad. 14-bit has 2^14 = 16384 baskets of amplitude, with the positive part of the signal covered by half that number, so 8192, which is close to the lower-level 16-bit example of 9830.

So APART from noise concerns, is it true that recording at a higher level will :

  1. Give a more accurate recording than a lower level?
  2. With 16-bit, lower the risk of sounding poorly like the 14-bit CDs did?

(I know using 20, 24, or 32 bit would make these questions irrelevant, but I’m limited to 16. Also, I’m very interested in the answers for technical understanding.)

Thank you for sharing your expertise.

2

I’m going to side-step the question with the idea that you should always record a live performance as loud as you can with no possibility of overload damage. Overload or clipping is almost always fatal.

The limit of 16 bit recording is something like 90dB, which, given the converter can do it, gives you 20 dB overload headroom and 70dB noise floor. More than enough for non effects-heavy production.

Chances are, the electronics before the A to D converter are worse than that and the vinyl record will certainly be worse, the restriction on quality will not be the 16 bit spec.

If you’re going to burn your vinyl to Music CD, the show is going to get converted into 44100, 16-bit, Stereo anyway.

Koz

3

Chris, you followed up with me on your above questions in a PM.

Probably Steve is the best person to go into details, but Koz is broadly correct that if you are recording vinyl the noise floor is likely to be above the -96 dB level anyway, and that if you had 24-bit recording capability (Audacity on Windows does not) the mere fact you were exporting to a 16-bit format would make the question fairly academic.

If you were recording 24-bit to a 24-bit format, it allows you to represent samples between -97 dB and -145 dB, the latter value being lower than most people can hear. 16-bit cannot represent those samples. 24-bit will give you smoother steps from one sample to another, but analogue playback smooths the more abrupt jumps in 16-bit anyway.

“Baskets of amplitude” is a strange term and I think it may be misleading you into thinking a 16-bit value at 0.331309 is inherently less accurate than one at 0.931309. I would see it that the lower level recording is equally accurate as a representation of the “noise and signal”, but it’s a less accurate representation of the possibilities afforded by the signal because the signal is smothered by the noise.

As for 14-bit/16-bit, yes 16-bit is more accurate than 14-bit in the same way 24-bit is more accurate than 16-bit.

I’m shutting up now because this isn’t my area of expertise.


Gale

4

though I don’t think there is much need - I think that Gale and Koz have covered the question thoroughly.

The only thing that I might add is that while 14 bit audio is not “audiophile” quality, it is not as bad as many people assume. Even a 12 bit recording (using the full 12 bits and assuming that the bit depth is the only limiting factor to the sound quality) sounds a lot better than what a lot of people listen to on their iPods. Uncompressed 14 bit audio is measurably better than an MP3 at 128 kbps. Audio does start to sound noticeably “rough” and noisy below 12 bit, which is why it is important to have a reasonable recording level. For recording from a source with a predictable peak level, (for example when transferring tape or vinyl to digital), a target peak level of around -6 dB is usually recommended (which works out as equivalent to 15 bit).

People often get “hung up” on the bit depth question, forgetting about other parts of the signal chain. For best quality recording it is important to keep as close as possible to an optimum level throughout the signal chain - so if recording with a microphone, that means using a microphone with low noise and suitable sensitivity for the sound, correct levels in the pre-amp stage (and a suitable pre-amp to match the microphone signal), the level through the mixing desk (if used), a sound card input that has a line level input level that is matched to the pre-amp/mixing desk output and finally the recording level in the computer/recorder. The recording level (hence number of bits available) is only one level in the chain, and as long as it is set to a reasonable level it is probably going to be a less significant limitation to sound quality than other parts of the signal chain.

Which gives you a very simple signal chain to work with. You have only one level to worry about. Put on a really loud record and set the volume level so that you are getting a peak level in Audacity of around -6 dB. You should then be able to go through your records without needing to adjust the level on the turntable. Note that the peak level for a 45 will not be the same as for an LP, so you may need to “re-calibrate” the setting when switching from 45 to 33 1/3.

5

How does one reliably do that?

I am accustomed to viewing the waveforms against the 0, .5, 1.0 scale immediately to the left of the waveform display, and then adjusting level up or down with the input level slider next to the microphone symbol.

That scale is not finely demarcated–from zero to 1.0 is about 1 vertical inch on my screen.

I normally try to get peaks to be above .5, maybe .8, but never 1.0.

Is that sufficient care/technique or should I be using some other tool/scale for level setting, such as the “mixer board” from the view menu, which I have studiously avoided so far? I have no idea what it does and haven’t examined it. I notice it has a DB scale, unlike the standard scale to the left of the waveform.

Suppose I record with a level clearly “too low” or clearly “too high”. If I then use normalize or amplify to alter the capture before export, will that adjust my poorly set levels to proper level, just as if I had set levels correctly to begin with–with no ill effects at all?

If that is so, why should I pay much attention to levels IF I can “correct” them after the fact (before export) with normalize, amplify, envelope tool, etc?

I come from analog land (low levels mean background hiss on playback, high levels mean clipping) where bad levels cannot be undone and I remain a bit unclear on how, if at all, a bad level can be corrected before export in digital land.

6

ignatz,

you can resize the meters by clicking and dragging, this makes them much more useful I have mine as a pair stretched across the whole width of my Audacity window - and IIRC Koz has each meter stretched across his screen witdth. I also have mine set to dB rather than liner - I find that more useful. And with 1.3 the settings you make are “remembered” in your preferences, so the next time you fire up Audacity you will get those some settings for the meters.

When I record LPs (and indeed also FM off-air) I aim for a max level of around -6 dB. Unlike Steve’s one-off setting suggestion, I recalibrate/reset for each LP. With experience you can learn to see from the vinyl groove patterns which are the louer/loudest parts of the recording.

Note that the -6db level can easily be exceeeded with a big click/scratch/crud on the record. Note that if that happens you can reset the peak-level in the recording meter by clicking in it.

After recording and editing my final processing step prior to multiple export is to Amplify (not Normalize) to -2 dB

WC

7

WC:

OK, I see what you mean by resize, but I don’t follow the “across the whole width of my Audacity window”. Attached are 2 pictures. One is the view I have always used. The other is zoomed to an extent. I don’t see how the zoomed is an improvement as you cannot see the peak.

Can you post a pic of what you mean?

How do you read “-6 DB” from a scale extending from 0 to 1?

I once saw a Steve post that gave the correlation between the two, but I cannot find it.

Do you use the mixer board, where you can read in DB directly?

Can you comment on my other points in my post–mainly about altering level AFTER capture with amplify or normalize. If that “works”, why bother setting to -6 DB? Why not use any old level and then alter before export?

I’m clearly not understanding something here and need more clarity.



That’s my practice also. I have transferred thousands of songs to analog tape over the last 40 years. I always set a new level every time something new landed on the turntable. I’m invariably familiar with the source and would just play maybe 15 seconds of it to adjust level, and then re-cue and start recording.

I’m aware that clicks or any unanticipated peak can push the level up, but am unclear on your second sentence. If I discovered something that drove levels too high, I would just move the input slider down a bit and don’t follow what you are saying in that second sentence.

levels zoomed.jpg
levels zoomed.jpg800×351 63.6 KB

levels normal.jpg
levels normal.jpg800×363 97.3 KB

8

Sorry - we’ve cross-posted here, so the topic has gone a little out of order.

For USB devices on Win XP or Linux you would normally have the recording level set at maximum, then adjust the level of what you are recording (the turntable volume control). I’m not sure what you do on Vista or Win 7.

Use the red recording meter http://manual.audacityteam.org/index.php?title=Meter_Toolbar
The meter can be dragged with the mouse and stretched to full screen width which makes it easier to read accurately.
You can also set the display range for the meters in “Edit menu > Preferences > Interface”. Setting it to a range of -96 dB is a good general purpose setting, but setting it to -60 dB will give you a bit more detail at the high end of the scale.

No. You can adjust the level of the recording, but it is not without ill effects.

If you record at too high a level you will get distortion - the tops of the peaks will be clipped off flat. If the clipping is bad enough to be noticeable when you listen back to it, then it is probably too badly damaged to effectively repair. Clipping distortion is a show stopper - delete and record again.

If you record at too low a level, then you will not be taking full advantage of the available 16 bits from your sound card/audio device. This is less serious, but means that the base noise level will be higher than it should be. This effect starts to become noticeable if you are -12 to -24 dB below 0dB (depending on how “clean” the audio is that you are recording and how critically you are listening. A peak level of -6 dB is recommended because it allows a reasonable amount of headroom while keeping the noise floor close to minimum. For live (microphone) recording it is generally advisable to allow more headroom (say 12 dB) to allow for unpredictable peaks.

It is better to record too quiet than too loud but should be recorded at a reasonable level.

9

There is absolutely no harm in doing that, though if using a USB turntable it may be more work than really necessary.
The assumption in not recalibrating for each LP is that the audio level across multiple LPs is close enough to fit comfortably within a reasonable recording level range - that is with a peak level on the quietest LPs at least -12 dB, and the maximum peak level on the loudest LP still below 0 dB.
If you have LPs that lie outside of this range then you will need to recalibrate.

Question for you vinyl guys - does it matter if the click from a scratch is digitally clipped at 0 dB in your recording?

10

Steve:

Thanks for the comments.

In an inexplicable brain fade, my previous posts would have you believe I was unaware of the red meters. Not true, although I did not know they could be dragged screen wide.

Possible explanation for my brain fade:

In analog land (cassette/open reel/turntables/1977), levels are critical and I am all over those red meters in that situation.

But I have been led to believe that in digital land (Audacity/WAV/mp3), level settings are not nearly as critical.

Accordingly, I have been just setting levels roughly by eyeballing the wave form peaks against the 0 to 1 scale on the left–getting peaks above .5, but clearly, I say never, above 1. I pay some but minimal attention to the red meters and figure that if random song A has a different level than random song B, my playback volume control app (mp3 Gain) will iron it out reasonably well. I know from experience that equal levels on meters doesn’t necessarily translate to equal playback “loudness”.

Have I been led astray and are levels as critical here as on open reel/cassette??

At times, I will normalize or amplify after capture, but typically I DO NOT. My normal practice is to just capture at a reasonable level, export as mp3, and then use mp3 Gain to alter tags and thereby volume on playback.

Bad technique? Have I been misinformed about levels not being particularly important within reason?

11

That all sounds fine.

Correct - not as critical, but while in the land of 16 bit a degree of care must still be taken.
Once it’s into a 32 bit (float) Audacity track, then the levels become extremely non critical - you can safely amplify down to -30 dB and back up again with no noticeable affect - you can’t do that on analogue tape :wink:
Note: Only later Beta versions 1.3.x can store values above 0 dB - the 1.2 series cannot.

In practical terms that should be close enough. “Never above 1” is the golden rule. Around about 0.5 (-6 dB) is a good guideline.

12

Am I “in the land of 16 bit” and must take “a degree of care”?

Or am I “into a 32 bit (float)” where…“the levels become extremely non critical”?

32 bit float is shown in my edit/preferences/quality/default sample format window, but I don’t pretend to understand that, other than since 32 is a larger number than 16, it must be better.

So I really don’t know which numerical land I am in.

13

You are in the land of 16 bit while you are recording (and when exporting - assuming that you are using a 16 bit file format as the destination)
If you have the default “Quality” setting to 32 bit float (Edit menu > Preferences > Quality), then once the audio has been recorded you will have arrived in the land of 32 bit float, and will remain there until you Export your completed project.

Very much better - when processing audio, the computer can calculate with fantastic precision.

Most consumer level sound cards and audio devices work at 16 bit.
Most high level audio devices are also capable of working at 24 bit, but unfortunately Audacity on Windows is not currently able to access 24 bit data from the audio device due to limitations in the current version of Portaudio (which Audacity uses to access the sound card). I believe this issue is being addressed by the Portaudio developers, so hopefully 24 bit sound cards will be fully supported in the future. I think that 24 bit audio is already supported on Linux and Mac, but I’ve not been able to test that.

14

It doesn’t seem to. Brian Davies’ ClickRepair seems to clean them up nicely even when they clip like that.

What I do is:
record one side of an LP (I usual mark big clicks or click section with temporary labels)
Export to WAV file for CR processing (leaving the original track in place for later comparison)
Process the WAV through CR
Import the CRed WAV into the open Audacity project
Check that the repaired track has been fixed properly by CR (mute the original track - use the clipped points in that track and the temp. labels to identify the places to audition)
Mandraullically repair any that are not fixed ok by CR (an extemely rare occurrence - usually longish time duration noise events)
Delete the original track

WC

15

ignatz,

you can reset your track waveforms to be calibrated in either linear or dB (or you can choose a couple of spectrum views - but I would not use thoise for recording mode as they are too slow to draw). Jist click on the “little-black-downward-pointing-triangle” at the top of the track control box - the third block lets you choose the track display format.

For the meters: either right-click in the meter or click on the meter’s “little-black-downward-pointing-triangle” (next to the mic and speaker icons).

Personally I prefer to work with my track display set to linear waveform - but with my meters set to dB display. It is the meters that I use to monitor the levels rather than the graphic waveform display.

============================================

One of the biggest differences in moving from analog-tape-land to digital-land is that with good tape it was possible to temporarily go into the red on the meters to get a good fat signal - in fact one was encouraged to do this. The physics of the tape and its recording technology meant that this could be done without damaging (clipping) the signal. However in digital land clipping is always bad, so getting to 0 dB is to be avoided at all costs beacuase it always damages the signal and usually sounds horrible. And as Steve says working in 32-bit float gives you plenty of headroom for later amplification.
WC

16

WC:

Can you comment on Davies ClickRepair declick and decrackle workflow.

I am currently doing this:

Open track in ClickRepair.

Set decrackle to zero.

Set declick settings as desired

Process and save as WAV. This yields a WAV file with CR extension that has been declicked but not decrackled.

Open the new file with the CR extension.

Set declick to zero.

Set decrackle setting as desired.

Process again and resave.

This yields a new file with a CR-1 extension that has been both declicked and decrackled.

Use the CR-1 WAV for further processing in Audacity.

I know you CAN do both in a single pass, but does that do declick first and decrackle second, as if in separate passes? I’m guessing not.

I know the preferred order is declick, then decrackle, but does that necessarily mean one, save, then the other? I’m guessing yes.

17

ignatz,

personally I never use the decrackle (most crackling on LPs comes from cold-pressings and I was always very fussy about taking those back to the shop - the record shop was very pleased when I moved to buying CDs …).

But knowing Brian and his meticulous approach to this I suspect that it is quite safe to do both in a single pass. What I woul do is find a clicky and crackly record - and with a small noisy section experiment with single and double pass and then compare the results - you will need good speakers or studio quality headphones (but I suspect you have those …).

Does Brian’s manual have anything to say about preferred workflow?

The other approach would be to email Brian asking on his opinion on best use of the software - I usually found him helpful when I have had queries in the past.

If you do this I would be extremely interested in your conclusions.

WC

18

WC:

I believe he states somewhere that the preferred workflow is:

Derumble, declick, decrackle, debuzz, and then dehiss.

Which in his world would mean Denoise LF, Click Repair, Click Repair, Denoise LF (I think, for Debuzz), Denoise. The only question is–how many passes of Click Repair–one or two? Can you DeNoise LF the rumble and buzz in a single pass–even though that would violate the above preferred order of operations?

But I did not see a single word implying that both can be done in a single pass–so I have chosen to do them separately.

I am from mono-land (pre 1960 recordings). When you get back to pre-1955, most original issues (back into the 1930s) were in fact on 78—which are quite prone to crackling–that semi-constant “frying bacon” sound not typically found on an LP.

So I am interested in decrackle to the extent it works.

Suppose you had a 1984 reissue of 1940s recordings. You would not consider decrackle?

I do know that some LP reissues were in fact mastered from actual 78 discs and these do contain the crackle–it’s just a form of groove noise, even on a mint 78. I’ll admit I’m not sure how that translates into a purpose for decrackle, considering that the reissue is on LP vinyl, not 78 shellac.

Tape was not typically used for original studio recordings until the late 40s. Ampex recorders I think. Anything earlier was recorded directly to disc of some type.

Considering your name, I imagine you know all of this.

I get quite a few mp3s that someone has personally made from their own 78 collection–and decrackle is useful.

I emailed Davies a couple of weeks back about lacking a “zoom into the waveform” capability. He did write back, confirming it did not exist, so I may write him on this issue.

My 21 day trials are about to expire. I think I will buy Click Repair and pass on Denoise. It seems Audacity is as good for denoising and it is certainly more user-friendly in that you CAN zoom into the waveform and make that 1/8 second of lead in groove an inch wide and therefore easily grabbable with the mouse.

19

I no longer have any 78s - my parents threw them all out when I left home to go to college :angry:

When I was doing my LP transfers I had a couple of blues compilation records that were taken from original 78s and long before the days of digital processing. so these sounded like 78s. I didnt use de-carckle on them - but I did experiment with Brian’s De-Noise. It worked ok, not as easy to setup as ClickRepair. I didn’t bother buying DeNoise as those were the only albums I had like that - but CR is an extremely valuable item in my e-toolbox.

I don’t bother with low frequency filtering as frm looking at the spectrum analyses my TT dosn’t seem to be prone to rumble. But if I did I would just use Audacity for this rather than Brian’s De-NoiseLF

The grey hair and grey beard are a bit of a giveaway too :slight_smile: - “all” is stretching it a bit far - but I have learnt a lot more from the last couple of years that I have spent on this board and helping out on the Audacity manual.

WC

20

WC:

I have another question or two about Click Repair, but will add them to your thread devoted to that app rather than continue here.

Please take a look there shortly.