Vinyl LP Digitization - USB and Bit Depth

Preparing to digitize a half dozen vinyl LP’s (can’t find them in digital form; other than maybe on Youtube). Read the "Sample workflow for LP digitization, and have a couple questions.
https://manual.audacityteam.org/man/sample_workflow_for_lp_digitization.html

The document says to record in 32-bit. Does that matter if the USB device is 16-bit?

When recording from a USB device does the “signal” remain digital all the way into Audacity? Or does it get converted to analog, or change bit depth, before Audacity sees it and coverts it back to digital? The reason for this question is related to the first question re: bit depth. i.e. if “signal” remains digital and is only 16-bit then recording at 32-bit would seem to be wasted space.

Thanks

Update:
I should add. There is no post processing to be done. Save the recordings as is. Levels will be set so no normalization etc. will not be necessary. Maybe a couple click removals but that’s all.

No, that doesn’t matter. It is recommended to record in 32-bit regardless of the recording device.

The USB device converts the analog signal to digital. It remains digital all the way into Audacity (and all the way to the hard drive, and back out again on playback until it reaches the Digital to Analog Converter (DAC) in the playback device / sound card).

There are several benefits to recording as 32-bit float:

  • Modern computers handle 32-bit float more efficiently than they handle 16-bit integer.
  • Audacity works internally in 32-bit float, so you avoid unnecessary conversions inside Audacity.
  • Any processing that you do in Audacity is much more accurate when the audio is in 32-bit format.
  • If you inadvertently go over 0 dB during processing, there is no damage provided that (a) the audio data is 32-bit float format, (b) you correct the amplitude (amplify below 0 dB) before you export.

The main down side is, as you say, it requires double the disk space while you are working on the project. However, if you are so critically low on disk space that that is a problem, then you are likely to run into other problems such as random crashes, severe disk fragmentation, poor disk I/O performance, etc. Disk space is cheap these days (even SSD drives are pretty reasonable), so disk space should be a non-issue.

There will not be any amplitude post processing. Maybe a few click removals. But that’s all. Disk space is not low. But don’t want to consume it for little to no benefit.

So if the source is 16 bit that would be the same as if CD rip. I don’t see benefit to 32-bit if no amplitude post processing. I’m I missing it?

My advice is leave the 32-bit floating-point default. The conversion from 16 or 24-bit to 32-bit float, and back is lossless. The ONLY “downside” is the TEMPORARY need for more disc space. Virtually all audio editors & DAWs use 32 or 64-bit floating point “internally”. Don’t try to out-smart the smart people! If you are not doing any “processing” you don’t need floating point but there’s also no need to mess with the defaults.

If you don’t want the data altered at all turn OFF [u]dither[/u]. As a practical matter, it won’t matter either way… You can’t normally hear dither (which is very-low level noise) or the effects of dither at 16-bits or better, and because vinyl is “self dithered” from the vinyl noise.

Levels will be set so no normalization etc. will not be necessary.

Digital volume adjustment is also HARMLESS. It very difficult to get the “perfect level” while recording without clipping and you might have to record twice (or 3 times). The standard practice is to shoot-for -3 to -6dB, leaving plenty of headroom. You don’t really need headroom… Nothing bad happens when you get close to 0dB but you’ll get clipping if you “try” to go over and analog levels are always unpredictable. (Pros typically record at -12 to -18dB.)

If you are old enough to remember analog tape, you wanted a hot signal to overcome tape noise, and tape tends to soft-clip so it was OK to go occasionally “into the red”. With digital there is no tape noise and it will hard-clip if you “try” to go over 0dB.

I generally normalize the album as a whole, keeping loud songs (relatively) loud and quiet songs (relatively) quiet.

Even if you normalize (AKA “maximize”) vinyl it’s usually quieter than a digital original and your digitized vinyl will be quieter than most of your CDs. There are two reasons for this. Modern CDs & remasters are often victims (or victors) of the Loudness War. Plus, the vinyl cutting & playback process tends to make some peaks higher and some peaks lower (without changing the sound of the dynamics). So if the digitized vinyl and digital original are both normalized for 0dB peaks, the digitized vinyl will have a lower average (or RMS) level and it will be (slightly) quieter. This increase in peak-to-average ratio (crest factor) fools some people into thinking the CD is more compressed, even when the LP & CD were made from the same master.

Maybe a couple click removals but that’s all.

Audacity has Click Removal (automatic), Repair (manual), or you can zoom-in and manually “re-draw” the waveform. [u]Wave Corrector[/u] is a FREE automatic declicker. [u]Wave Repair[/u] ($30 USD) offers several click removal methods and it only “touches” the audio where you identify a defect (which is usually very time consuming).

For hum, hiss, and low-level crackle, you can try the regular Audacity Noise Reduction effect. But, you can sometimes get artifacts so listen closely to the results and you may only want to apply it to fade-ins, fade-outs, or other quiet parts where the background noise is noticeable.

Thank you for that. Very informative and good things to think about.

I don’t typically normalize my ripped CD’s either. Only if they are overly quiet. One CD was done that way intentionally to maintain wide dynamic range and have the drums really snap instead of being compressed. But makes if difficult to listen to in a noisy environment like a car, or shuffle mode.

I think the vinyl record levels will be fine without normalization. I’ve digitized these once before eons ago when mp3 was relatively new and disk space was premium. So only kept a 128 kbps mp3. Plus it was analog line in source to an IBM ThinkPad that required a special line level adapter to use with the mic input. So figured I’d better do these more proper like while the turntable still works. It’s getting old.

Anyway I select a loud section to use for level setting. Then there’s no need for normalization or even much room for it. Did test pass with an album that won’t be digitized and there was only 1.2 db of head room for normalization. Of course could compress it. But don’t want to go altering the original when not warranted.

Again, thanks for your thoughts. Really appreciate it.

P.S. Oh yes I do remember the tape days. Used tapes of all the vinyl to save the vinyl.

Here’s how I see the bit deepening.

16-bit source (USB A/D)
Audacity record/capture as 16-bit
Export (archive original) as 16-bit wave

If later decide to do amplitude change processing.
Load the 16-bit wave archive original into Audacity as 32-bit float.
Export as 16-bit wave with dither, or as 32-bit without dither.

  • Or -

16-bit source (USB A/D)
Audacity record/capture as 32-bit float
Export (archive original) as 32-bit wave, or 16-bit wave (dither would seem pointless if no amplitude change processing has occurred)

To summarize. If the source is 16-bit then why not archive as 16-bit? Seems pointless to store 16-bit integers as 32-bit float values. Can always load it into Audacity as 32-bit float for amplitude change processing if desired.

Well those are my thoughts. What are the flaws in this?

A potential flaw could be if there is any bit depth change between USB input and Audacity capture. Would hope there wouldn’t be, but I don’t know that to an absolute fact.

Thanks

What I’d do is:

16-bit source.
Record as 32-bit float.

If, after recording, I feel it is necessary to Normalize, or apply a bit of EQ, or fade out the end, or anything else, then do that and export archive recording as 16-bit with dither.

If after recording, I don’t feel it is necessary to do anything, then export archive recording as 16-bit without dither.

If I later want copies, I can just copy the file (no need to even open Audacity).

Of course, you are free to do whatever you like. :wink:

Of course, you are free to do whatever you like.

Well naturally. :wink: But whatever I like can be changed for a warranted benefit.

Well my process would be to archive the unaltered recording/capture. Any processing for usage copies will be derived from that.
Similar to treating a CD rip as an unaltered archive that is loaded in to Audacity for processing (as 32-bit float if amplitude processing). But the original (archive) is never changed (other than maybe adding some basic metadata; artist, album, year, track #, track title, publisher/label).

I don’t typically normalize my ripped CD’s either. Only if they are overly quiet. One CD was done that way intentionally to maintain wide dynamic range and have the drums really snap instead of being compressed. But makes if difficult to listen to in a noisy environment like a car, or shuffle mode.

Normalization doesn’t change the dynamics. Dynamic compression reduces dynamics.

That quiet CD may already be 0dB normalized. Peaks don’t correlate well with perceived loudness. A normalized highly-dynamic recording will usually sound quieter than a normalized highly-compressed recording.

A normalized highly-dynamic recording will usually sound quieter than a normalized highly-compressed recording.

Exactly. The CD was perceptually soft because of the high dynamic range. Though if I recall it still had several db of head room. But even that was not enough to create a reasonable balance with others CD’s and for noisy environment listening. Ended up compressing it enough to get about 15 or so db. It could probably still use about another 5 db to be level with most other CD’s. But those drums are going to start loosing their crisp snap at some point (at 15 db compression it is already becoming noticeable). Don’t want to alter it anymore than necessary for reasonable target environment use (mostly car road trips).

-6db is giving up about half the available resolution (1 bit). Maybe not so significant for 24-bit as there is still 23-bits of resolution. But reducing 16-bit resolution to 15 is more significant. For 24-bit or greater I’d probably be more lenient. -3 to -6db would be no big deal.

For digitizing audio that has already been tamed, vinyl for instance, at 16-bit resolution, I shoot for peak better than -2db without clipping. Utilize as much of the available resolution as possible.

Pros typically do this… is meaningless without context (live performance, studio recording session, etc.). Here we are dealing with tamed audio (professionally mass produced vinyl). -12 to -18db would just be silly for this purpose.


Hot signal to tape was about the inherent media noise (tape hiss). Not the upstream signal path noise.

Though digital does not have that inherent media noise, the principle to apply gain early as possible in the analog signal path still applies, to avoid including noise unnecessarily in the recording. Regardless of the recording media.