RECORDING at 16 bit vs internal 32-bit float?

I have tried searching, but I either cannot find the right info or I am just too dense to understand this. The documentation, and many posts on this forum, clearly state that Audacity uses 32-bit float for internal processing, and it is strongly suggested that we not change that in the Quality settings. I got that.

BUT, the same documentation says that in most causes Audacity cannot actually record at anything higher than 16-bit (and almost no analog to digital converters sample at higher than 16-bit anyway), so it seems really confusing that we need to always use 32-bit float if the actual audio is only at 16-bit.

I am not asking for a detailed explanation of the internal workings, but in a real world situation, how does this apply? For example, I routinely record audio on two different computers, and then do editing and sound cleanup on a third computer:

• An computer A, I record streaming audio which is generally modest quality, but rarely requires any editing other than splitting files and normalizing sound levels.
  
  
  
• On computer B, I record LP and tape conversions. These files should be much higher quality, and also often require a lot of post processing.
  
  
  
• I move all the recorded .wav files to computer C for any editing and final library management.

So should I be constantly saving ALL recordings in .wav 32-bit float format until I have completed final editing, or are there some lower-quality files where it just will not make any difference at all?

Finally, if a save a final recording as a 16-bit 44100 format FLAC file, and then at some later time decide I want to do some additional editing, why should I then let Audacity re-convert that same audio back into 32-bit float? My uninformed logic just says that would be a pointless conversion.

Firstly, Audacity’s default settings of using “32-bit float” for tracks, and “16-bit” as the default for WAV export, works very well in the vast majority of situations.

One of the main benefits of 32-bit float is that it can go beyond 0 dB, whereas 16-bit and 24-bit (“integer” formats) have an absolute limit of 0 dB.

Try this:

1. Import an audio track

2. Set the track format to 16-bit (See: Audio Track Dropdown Menu - Audacity Manual)

3. Apply “Amplify” with “New peak amplitude” set to “6.0” (positive 6.0 dB) and “Allow Clipping” enabled (ticked)

4. Normalize to - 3 dB (negative 3.0 dB)

5. Observe that peaks have been “clipped”, and on playback it sounds horribly distorted.
   
   
   Now with 32-bit float:

6. Import an audio track

7. Ensure that the track format is “32-bit float”

8. Apply “Amplify” with “New peak amplitude” set to “6.0” (positive 6.0 dB) and “Allow Clipping” enabled (ticked)

9. Normalize to - 3 dB (negative 3.0 dB)

10. Observe that the audio track has not been damaged and it sounds OK.
    
    
    There are additional benefits to using 32-bit float, but they are more subtle. Processing is more accurate (higher quality) and a bit faster when using 32-bit float.

BUT, the same documentation says that in most causes Audacity cannot actually record at anything higher than 16-bit (and almost no analog to digital converters sample at higher than 16-bit anyway),

Many soundcards and most “good” USB audio interfaces are 24-bits, although the rumor is they usually aren’t accurate to more than 20-bits.

Just for reference the “pro studio standard” is 24-bits/96kHz. In most cases I’d say that’s overkill, but it’s the the only downside is larger files so if your hardware support it, “why not?”. And, when you’re done you can export to the format of your choice.

Then, virtually all audio editors/DAWs work in floating point ( 32-bit or 64-bit) internally. The conversion from 16 or 24-bit integer to 32-bit floating-point and back is lossless.

so it seems really confusing that we need to always use 32-bit float if the actual audio is only at 16-bit.

In addition to what Steve said, digital signal processing is actually “easier” in floating point.

There is no point in increasing the sample rate. Some people claim that some effects work better at higher sample rates but I’m skeptical. And if that’s true the software developer should design the effect to temporarily up-sample during processing.

So should I be constantly saving ALL recordings in .wav 32-bit float format until I have completed final editing, or are there some lower-quality files where it just will not make any difference at all?

Not necessarily. As long as you haven’t done anything to push the peaks over 0dB you can export to whatever format you’re recording in. i.e. If you’re recording at 24-bits it’s usually best to keep the “maximum quality” until the end where you may want a different format.

But as a practical matter, “CD quality” (16-bit, 44.1kHz) is better than human hearing so you probably won’t hear a difference so if you save your intermediate files as 16/44.1, or if you have a 16-44.1 file and decide go-back and re-edit it later, that’s OK.

I really appreciate the answers here - I used to think I knew most of this stuff, but I am getting kinda old, and I cannot tell the difference between what I have forgotten and what I just no longer think might be true! While I had some reasonably high-end kit back in the 70s, I know all that stuff is pretty much junk these days, and I am not sure how much I can actually hear anyway. I do not do any recording or editing at a professional level; it is just for my personal use and enjoyment. Mostly just trying to preserve some very large archives of pre-digital music.

I know I am probably way over-thinking this stuff and going overboard on trying to preserve as much quality as I can, but I got kinda paranoid when I had to re-record numerous things due to rookie mistakes and failure to actually understand the basics of ADC back 20 years or so. As a result, now I am afraid I won’t have the best source material next time I think I have to start over on something. And frankly, most of my questions and concerns would not even come into play if I just used one computer to record, edit, and then save the final product in 16-bit 44.1kHz files. But I got kinda wrapped around the axle in concerns with multiple conversion and generation losses from moving files around during processing.

Your responses have helped, along with my finally having stumbled upon the “sample_format_bit_depth” page in the Audacity Documentation. To calm my fears about format conversions, I am now using the new 3.0 project format .aup3 files for all recording and intermediate storage of files until I get ready for final export. But now for one last question. This seems rather obvious to me, but I thought I would ask for opinions just in case I am not as smart as I think I am:

I have always saved the final results of my recording/editing efforts in 16-bit 44.1kHz .wav files, just as an archive in case I wanted to do additional editing or QC comparisons at some point in the future. But I have recently found that Audacity can process my .flac files just like it does the .wav files, so I am now assuming that there is absolutely no benefit from ever keeping the archives in larger .wav format instead of just using the same Audacity export in .flac. Is this a correct assumption?

I am now assuming that there is absolutely no benefit from ever keeping the archives in larger .wav format instead of just using the same Audacity export in .flac. Is this a correct assumption?

Correct. FLAC is lossless.* In addition to getting a smaller file tags/metadata are better-supported with FLAC. Maybe sometime in the future FLAC will “go out of style” but since it’s lossless you can always convert to another lossless format as long as a decoder is available.

I am now using the new 3.0 project format .aup3 files for all recording and intermediate storage of files until I get ready for final export.

It’s still a good idea to make a WAV file immediately after recording as a back-up. AUP3 files should be more robust than previous AUP files but I’m already seeing a few posts about “lost data” or “can’t open file”. WAV (or FLAC) files are “simpler” and are less likely to get messed-up.


\

• FLAC doesn’t support floating-point but you’re probably not going to archive floating-point anyway.

Not likely in the foreseeable future. It’s been the most widely used lossless compressed audio format for many years, and hardware / software support for FLAC is still growing. Even “Groove Music” (the default music player for Windows 10) supports FLAC.