24- to 16-bit Workflow

Sorry if a noobie question, but I could not find this already answered anywhere.

I archive analogue LPs and tapes to CD. For many years I’ve used a CD recorder (part of my hifi rack), so I’m going directly to CD format (16-bit). I sometimes use Audacity later to cut out gaps and insert track breaks, as it’s easier and more precise than doing these on-the-fly during the live recording, but I don’t do any other processing. This works fine, but the lack of headroom with 16-bit forces me to push recording levels very near to 0db (I usually try to peak at -3db to allow just a little leeway).

I am thinking of replacing my CD recorder with a solid-state recorder capable of 24-bit. This would allow me to set peaks to -18 to -12db - so no chance of going ever over 0db.

What I want to be able to do in Audacity is convert to 16-bit, as I still want the final result to be a standard audio CD, and for it to “map” the levels correctly. For example, let’s say I recorded an LP to 24-bit and I set my levels to aim for -18db peak. The actual peak during recording might have been -14db due to an unforeseen particularly loud drum beat or something. Ideally I’d like Audacity to scan the entire recording to find the loudest peak (-14db) and map that to 0db in the 16-bit output. This would give me a mathematically perfect recording level.

Can Audacity do this?
Would I lose quality due to re-quantisation doing this?
I assume my 24-bit master should be made at 88.2Khz so that at least there wouldn’t be re-sampling errors (you’re just throwing every alternate sample away to get 44.1Khz). Is this correct?

I like to keep recordings faithful to the original, so I don’t want to ‘muck about’ too much in the digital domain.

Many thanks for your help.

Ideally I’d like Audacity to scan the entire recording to find the loudest peak (-14db) and map that to 0db in the 16-bit output. This would give me a mathematically perfect recording level.

Easy! Audacity scans your file when you open it. The Amplify effect will automatically default to whatever (positive or negative) gain is required to hit 0dB, or you can simply enter a desired new peak value. Or, you can use the Normalize effect which allows you to normalize the left & right channels independently* and remove any offset.

This works fine, but the lack of headroom with 16-bit forces me to push recording levels very near to 0db (I usually try to peak at -3db to allow just a little leeway).

It’s up to you to leave headroom… The “ceiling” is the same… More bits give you a lower “floor”.** If you have 24-bits, go ahead and use 'em. But, there is no need to get close to 0dB with 16-bits. -6dB is fine, or even -9dB is OK.

With digital your recording levels are not that critical (as long as you don’t hit 0dB). Pros do often record at around -12 to -18dB (at 24-bits), but I think this is “traditional” to leave room for mixing & effects from back when Pro Tools used integer processing. Your ADC & DAC (and CD’s and “regular” WAV files) are still integer and hard-limited to 0dB, but most audio editors (including Audacity) use floating-point for internal/temporary processing so there’s virtually no upper or lower limit (while mixing or applying effects, etc.).

Would I lose quality due to re-quantisation doing this?
I assume my 24-bit master should be made at 88.2Khz so that at least there wouldn’t be re-sampling errors (you’re just throwing every alternate sample away to get 44.1Khz). Is this correct?

Don’t worry about it. The guys who do [u]Blind ABX testing[/u] have pretty-much demonstrated that nobody can hear the difference between a high-resolution and a copy downsampled to “CD quality”.

…No, you can’t just throw-away every-other sample because you can get aliasing (false frequencies). You have to low-pass filter before downsamling and every proper downsampling algorithm does that.

\

  • As a rule, I wouldn’t normalize independently because matching the peaks won’t necessarily balance the left/right average or perceived balance and you are just as likely to throw it off. The same goes for DC offset removal… If you don’t have a true DC offset (which you shouldn’t if your hardware is OK), don’t use it. It can sometimes introduce an offset.

** When you play a file with a bit-depth that doesn’t match your DAC, the data is scaled-up or scaled-down to match the hardware (by the drivers or software). So a 0dB 8-bit file is just as loud as a 0dB 24-bit file, even though the 24-bit file has way-bigger “numbers”.

Thank you DVDdoug for taking the time to reply. I understand all that you’ve said.

If I buy a new recorder, is there any advantage at all in mastering at a sample frequency that’s an exact multiple of the desired final format (88.2 down to 44.1) or could I master at 96Khz just as easily?

Just touching on my current recording practice of peaking at -3db - As I mentioned, I never previously changed levels in software after the actual recording. I only used Audacity to put in track breaks. Therefore the level I recorded at was the level I got on replay, and that can make for rather quiet CDs if you drop 6db or more. Particularly if the archival was for someone other than myself, they’d moan that the volume was too low. But I fully understand this issue goes away if you later normalise the volume level.

When I first got the CD recorder (1997?) the position was much worse because CD-RW hadn’t been invented. I was literally burning my final copy direct in real time. If I set the recording level wrong by too much, the disc was toast and I had to start again. Ditto if the record jumped - either you live with it on the recording forever or you start again. And the recorder has to take special expensive media designed for audio use (for copyright reasons this was enforced - you had to pay a tariff on blank media prices to discourage piracy). I now burn to CD-RW (still specialist audio ones, but at least they can be re-used), push it through Audacity, and then burn the final to regular cheaper CD-R.

Thanks again.

No, none whatsoever.
It is often said that using 88.2 kHz is better than 96 kHz in the mistaken belief that it can be resampled “more exactly” by either dropping every other sample, or taking the average of adjacent samples, or some other overly simplistic way. However, as DvdDoug pointed out, that’s not true. Down-sampling must be bandwidth limited to avoid aliasing. Without proper low-pass filtering, any frequencies present that are greater than half the new sample rate will “bounce back” from half the sample rate to create “phantom” lower frequencies. Resampling, when done properly, avoids this problem by filtering out frequencies that are above half the target sample rate.

You could master at 96 kHz just as easily, but even that is arguably overkill. At 44.1 kHz sample rate with modern hardware, audio frequencies up to around 21 kHz can be recorded accurately. Some children with extremely good high frequency hearing can detect frequencies up to nearly 20 kHz in ideal test conditions, though for adults the upper frequency limit is generally much lower. For real-world audio, extreme high frequencies are so low in amplitude that they are largely masked by other sounds. In short, 44.1 kHz is perfectly adequate for recording audio for human listening (some other animals can hear significantly higher frequencies than we can).

If I buy a new recorder, is there any advantage at all in mastering at a sample frequency that’s an exact multiple of the desired final format (88.2 down to 44.1) or could I master at 96Khz just as easily?

Practically speaking it won’t make any difference. Theoretically, you wouldn’t want to make any unnecessary conversions so you’d do everything at 44.1kHz if you were making a CD. (Most pro recording is 24/96 then it’s downsampled for CD mastering.)

Just touching on my current recording practice of peaking at -3db - As I mentioned, I never previously changed levels in software after the actual recording. I only used Audacity to put in track breaks. Therefore the level I recorded at was the level I got on replay, and that can make for rather quiet CDs if you drop 6db or more.

Again practically speaking, you can change the volume digitally (within reason) and it’s harmless. Professional audio productions go through multiple up and down volume changes throughout the production cycle. Mathematically there is (usually) some rounding but it’s tiny-tiny and you don’t have to worry about it.

But, if you are downsampling to 16-bits you are (obviously) loosing resolution and the lower the level the more resolution you’re loosing… So theoretically, it would be best to boost the volume to 0dB (or near 0dB) before downsampling. But that’s all “theoretical” and you’re not going to hear any quality loss. (Analog volume adjustments are also imperfect and if you’re going to boost the volume it’s usually better to do it digitally, earlier in the “chain”.)

If I buy a new recorder…

I assume you know, you can record with your computer. If you have a desktop/tower computer the line-input is often quite good, although it may be 16-bits and it can be hard to tell if you don’t have the “real specs” because the drivers will up-sample if you record at 24-bits and you’ll never know.

If you have a laptop with no line-in (or if you want to make sure you’re getting 24-bits) there are all kinds of higher-end [u]USB Audio interfaces[/u]. Just make sure to get one that has line-inputs/ (Switchable mic/line inputs are common.)

Brilliant, thanks for all the replies. I had overlooked the fact that I could record the 24-bit master at 44.1Khz and save re-sampling - doh!

I am aware I could use a PC/laptop, but I work with computers for a living and I like to get away from them as far as possible when I’m in hobbies time :smiley: It’s a personal thing but I like the aesthetics of a dedicated recording device sitting in my hifi rack. I’m considering the Denon DN500 (rackmount) or one of the Marantz portables which would give me the option of doing some field recording. My budget is £400 tops.