Re-quantising 24-bit to 16-bit

Hi, I do a little archiving of both records and analogue tape as a hobby. My final product is always a physical ‘Red Book’ CD, either for my own purposes or for friends/family who have cherished records they want copied.

I am contemplating starting to record in 24-bit and then re-quantising to 16-bit. The main reason is that I can be completely relaxed about setting the recording level. At the moment I feel I have to ride too close to 0dB so that I use “all the bits”, and so that the output doesn’t sound oddly quiet on replay. If the level then goes into the red during recording, I will more than likely curse and start all over. My understanding is that I could set peaks to -18 or -12dB in the 24-bit domain, and then normalise the levels and output to 16-bit, thereby taking all the stress out of this.

I have two questions about this:

  1. If the 24-bit master has a massive peak in it caused by a scratch on the record, won’t this skew the normalisation process? Should I care if a blemish (ie not part of the music) goes into the red? Can I tell Audacity to ignore it when it does normalisation? Should I try and edit the blemish out first?

  2. If the original source has a limited dynamic range (like an LP) then when I requantise to 16-bit, then I guess I just want to truncate the bottom 8 bits and discard them. Is this what Audacity does? (I realise it also may add some dither). However, if my source is a live band, the dynamic range could be enormous. What I ideally want to do in these cases is compress, not truncate. I’m making this up a bit - am I completely wrong?

Thanks for your help.

Yes, very likely.

Only so far as your vinyl has a scratch, but not that the scratch goes into the red.

Yes.
If there are only a few such blemishes, the “Repair” effect usually does an excellent job: Repair - Audacity Manual

Although vinyl has quite limited dynamic range (compared to CD), it’s not really a fair comparison to look at “noise floor : peak signal” as the dynamic range for vinyl. Sounds can still be audible, even when below the noise floor, and human hearing is very good at “mentally filtering out” noise when listening to a favourite record. The “effective dynamic range” is probably a lot more than a simplistic “peak : noise floor” measurement.

By default, Audacity is a bit more sophisticated than simple truncation.
Truncation will cause rounding errors for all sample values that are not exact 16-bit values, and these rounding errors create “quantization noise”. Quantization noise for 16-bit, is very low level, and only really audible when listening to very quiet sounds (such as the tail end of a fade-out), but nevertheless it can be avoided. Rather than simple rounding, a common practice (all studio do this) is to randomise the rounding errors in a carefully calculated way that focuses this noise in frequency ranges where human hearing is least sensitive. This is called “dithering”, and Audacity does this by default.

The effect of dithering is to replace quantization noise with (less obtrusive) dither noise. This can also increase the “effective dynamic range” to over 100 dB.

This is academic really, as the noise level from an album will be substantially higher than either the quantize noise, or dither noise, but reassuring that Audacity is giving you the best possible by default :wink:

The main reason is that I can be completely relaxed about setting the recording level. At the moment I feel I have to ride too close to 0dB so that I use “all the bits”, and so that the output doesn’t sound oddly quiet on replay.

There’s nothing wrong with 24-bits (as long as your hardware supports it), but even with 16-bits you’ve got lots of dynamic range and it’s really not necessary to use all of the bits. And of course, it’s common practice to amplify/normalize digitally after recording, and as you’ve said you’ll have to amplify if you record at -12 to -18dB.

  1. If the 24-bit master has a massive peak in it caused by a scratch on the record, won’t this skew the normalisation process? Should I care if a blemish (ie not part of the music) goes into the red? Can I tell Audacity to ignore it when it does normalisation? Should I try and edit the blemish out first?

You’ll have the same issue at 16-bits, but perhaps not as severe. If you are not going to fix it, it’s better if it’s clipped to limit the loudness of the defect. If you are using an automatic vinyl de-clicking application it might be easier for the software to detect the defect it it’s not clipped.

  1. If the original source has a limited dynamic range (like an LP) then when I requantise to 16-bit, then I guess I just want to truncate the bottom 8 bits and discard them. Is this what Audacity does? (I realise it also may add some dither). However, if my source is a live band, the dynamic range could be enormous. What I ideally want to do in these cases is compress, not truncate. I’m making this up a bit - am I completely wrong?

Not compress, dither. As Steve says “it’s academic”. At 16-bits and normal listening conditions you’re never going to hear the dither (or the effects of dither). And since dither is noise, vinyl & tape already have plenty of existing analog noise and are “self dithered”. You are not going to hear the very-slight additional dither noise.

archiving of both records and analogue tape as a hobby. My final product is always a physical ‘Red Book’ CD

I don’t think I would use the word “Archive” and CD-R in the same sentence. Many of my generally well-kept CD-R disks are discolouring and some of them no longer play. The burnable dye layer on a CD-R is a close cousin to color photograph dyes. That would be all the photographs in the bottom of the credenza with the serious fading and color shifts.

We’re told the high quality Sony CD-R blanks ceased production when the Fukushima Tsunami destroyed the factory.

All that is different from bought CDs which, someone will correct me, actually has little pits burned unto a robust aluminum layer.

Data storage is a lot more strenuous than people think as well. Keep valuable work in two different places and rotate it periodically.

My vinyl records still play perfectly.

Koz

Yes they are dyes, but they are probably a lot more stable than color phographic dyes. However a awful lot depends on the the particular CD-R stock, the storage conditions and probably just plain “luck”. Wikipedia puts the average lifespan of CD-R at 10 years, Some manufacturers claim 100 years. As and experiment I pulled out some 10-15 year old CD-Rs I had stashed away in a drawer and they read just fine.

The pits are actually moulded into the plastic. The aluminium layer is a uniform layer much like on a common household mirror. In principal they could last for centuries, but there are cases where the relatively thin layer of epoxy over the aluminium (the label side) has failed, resulting in the aluminium corroding and the disk becoming unreadable.

Indeed I would suggest keeping the collection on spinning (as on running) disk on your computer plus a backup in the cloud (eg AWS S3/Glacier) in addition to the red-book CD-Rs. The more copies in more places you have the more likely it is the data will survive.

My vinyl records still play just fine as well… But is often an adventure getting the turntable up and running properly again.

Thank you for the replies. I agree perhaps ‘archiving’ was the wrong choice of word. I copy to CD because it’s convenient, and I can’t hear the difference between it and the source. And for other people I copy to CD as they mostly don’t have turntables or tape decks any more - at least not as good as mine :wink: - Rega and Nakamichi respectively. I do make image files from all my CD-Rs, and my computer is backed up in multiple places. I’ve never had a CD-R go bad yet, but if I did I could burn another one.

I am sorry to appear thick, but I’m struggling to get my head round this. My goal is to make a 16/44 copy that is as faithful to the original source as possible. I don’t do any editing of defects or re-EQ-ing. So far I’ve been happy recording straight to 16-bit. There is nothing then needing doing to the file - no process at all which could affect quality. My only reasons for considering a new recorder are:
o The quality of a new recorder will probably be better than my 20-year-old CD recorder
o New recorders will record to SD card. My current recorder uses CD-RW which have a lifespan, have to be kept very clean, and may incur read errors when transferred to the PC
o If I buy a new recorder I may as well use 24-bit, set the recording level much lower and normalise later. This takes the stress out of setting recording level (I have been known to play a whole album side before recording to try and spot where the peaks occur to set levels accordingly - it’s very time-consuming).

So I still am unclear on the following:

o If I record at 24/44, normalise, and convert to 16-bit, is this process transparent or am I introducing losses compared to if I’d gone straight to 16-bit?

o On rare occasions, if I recorded live to 24-bit and the recording had a very wide dynamic range, how might I later compress that so that it is playable at normal listening levels?

Thanks again

If you “really” record in 24-bit, the process is better than recording straight to 16-bit.

Note that to really record in 24-bit requires that:

  1. the hardware supports 24-bit
  2. the drivers supports 24-bit
  3. both hardware and drivers are configured to use 24-bit
  4. PortAudio (in Audacity) captures in 24-bit or higher.

The last time I checked, Audacity could only receive 24-bit data when using WASAPI (or ASIO, but that’s not supported out 0f the box).

So you need:

  1. 24-bit hardware,
  2. set the device to use 24-bit in the Windows Sound Control Panel,
  3. Item (2) takes care of item (3)
  4. set Audacity to use WASAPI as the “host” in the device toolbar.

Note that item (4) above says “24-bit or higher”. The default in Audacity is to “record as” 32-bit float. This is highly recommended.

So your check-list is:

  1. Hardware
  2. Windows Sound Control Panel
  3. WASAPI + 32-bit float

“Normal listening levels” and “dynamic range” are different things.

For normal listening levels, fix any clicks / scratches, or at least reduce them so they are no higher than the loudest peak of “normal” audio, then Normalize.

To reduce dynamic range, you need to use a “compressor” or “limiter” effect. By definition, this affects (an “effect”) the sound, so it’s not an exact copy of the vinyl. There should be no need to apply dynamic compression to vinyl or tape recordings, as they don’t have particularly high dynamic range to start with.

Steve,

Thanks for the reply, but the PC hardware shouldn’t be an issue - I’m considering buying a standalone SD-card recorder, probably the Denon DN500R.

I thought most (all?) recorded music was compressed to some extent because it is assumed that we won’t be playing it back at the kind of volume that you’d have in a live gig. If you didn’t compress then, at comfortable listening volumes, I assume you’d struggle to hear quiet passages at all. You’d have to keep reaching for the volume control (effectively applying manual compression during replay by turning the volume up and down!). This isn’t necessarily anything to do with recording at 24-bit. 16-bit would have the same issue, albeit to a lesser degree. I didn’t think any commercial CD has a full >90dB of DR for this reason. I don’t record live venues often so this is a bit hypothetical for me, but I’m interested in the principles involved.

If you have any clicks, pops, scratches, dust at all then I would strongly urge yopu to look ta using a piece of software called ClickRepair. It was developed by a retired Australian mathematician when he was digitizing his vinyl - he started out with manual editing repairs but soon realized he could automate the process with algorithms. And a darn fine job of it he made too - I used ClickRepair with my vinyl transfers and it produced almost magical results. See this sticky thread: Click/pop removal - ClickRepair software


And if you are then really serious about getting the very best quality out of the vinyl, you should then let the LP cool down before re-playing it to record it. The act of dragging the stylus through the vinyl grooves heats up the vinyl and distorts it slightly - hence the need for the cool-down. But only ultra-purists do this :sunglasses: :nerd:

With a good eye you should be able to spot the louder parts of an LP’s groove - more activity in the vinyl in those parts.

WC

I wonder how much of that is the product of the legendary audiophile imagination (like “audiophile power leads”).
If you play an album twice, the contact point between stylus and groove has already had around 15 to 20 minutes to cool down between plays.

A standalone recorder would take care of it nicely, in which case all you need to do is use the Audacity defaults: 44100 Hz sample rate, 32-bit float sample format, shaped dither.
(the defaults were carefully chosen with sound quality in mind)

Yes. The the recording on the vinyl / tape will already be compressed, so generally there is no need to compress it more.

An exception is that if you wish to listen to music that has a large dynamic range (typically orchestral music) in a noisy environment (such as in a car/truck), then you may want to compress it further so that you can hear the quiet passages over the (engine / road / wind) noise. “Chris’s Compressor” was developed specifically for this situation.