Topic: 24bit madness

[jerryfreak]

turns out one of the bigger seeders in the scene was collecting 24 bit recordings for archiving. He was getting drives of shows from all the tapers that have been doing it for years, and found out that the majority of them that thought they were taping at 24 bit were actually recording at 16 bit + 8 empty bits for years and never knew it.

seems that a lot of the ones that actually had real 24 bit sources were the peeps i built taping laptops for from 2001-2005 (my friend who called me about it actually has one of these, i put about half a dozen of these rock-solid tested-to-death machines in the hands of  schoeps and B&K tapers in that period)

moral of the story, if youre building a computer taping setup, TEST TEST TEST! Test it for dropped samples under all sorts of demanding conditions. test it for resampling to make sure your interface is good. look at the resulting files in wavelab's bitmeter to make sure youre getting full 24-bit resolution. record something peaking at -90 dB and normalize it to make sure youre getting full resolution.

goes to show how insignificant of a difference 24 bit vs 16 bit actually is in a real recording environment. i always get a good laugh at all the torrents that are basically 'high end turntable>absolute garbage AD>24 bit rip'  

[ghellquist]

Cannot agree more.
I did a magazine review on a bunch of the boxes a few years ago. All of them proudly announced 24 bits. All of them had S/N (signal to noise-ratio) around 90 dB, well below anything into 24 bit land. The only effect of going from 16 to 24 bits on the same equipment is getting a lower noise floor. Well, if the noise floor of the analog circuits is high enough any gain is swamped by that.

Now, worth to remember for everyone is that 16 bits is plenty good when used well. Not a single CD ever sold has more than 16 bits and to me a lot of them sound good.

Gunnar

[JasonSobel]

look at the resulting files in wavelab's bitmeter to make sure youre getting full 24-bit resolution.

and for people who don't use WaveLab, there are a few free bit-meter VST plugins that will work in any audio program.
"Bitter" is a good one (do a google search for "bitter VST").
Solid State Logic (SLL) also has a good one called X-ISM.  this one can be downloaded here:
http://www.solid-state-logic.com/music/X-ISM/index.asp
You have to register with SSL to download this plug-in, but it's free.

[DSatz]

Not only would these recordings have only 16-bit resolution, but quite possibly they were also undithered. That would be the case if the signal itself was quiet enough, if the A/D was putting out > 16 bits and only the uppermost 16 bits were being stored in the 24-bit files.

[Gutbucket]

<snip>
there are a few free bit-meter VST plugins that will work in any audio program.
"Bitter" is a good one (do a google search for "bitter VST").
Solid State Logic (SLL) also has a good one called X-ISM.  this one can be downloaded here:
http://www.solid-state-logic.com/music/X-ISM/index.asp
You have to register with SSL to download this plug-in, but it's free.

Thanks Jason.  I've been meaning to look deeper into this, especially since reading Bob Katz's book and website, so I downloaded both meter plugins you linked above to compare them.  I'm not at my DAW computer, so I used a VST wrapper for Foobar to give them a quick look on this machine.  I played a few raw 24bit files of my recordings which I happened to have on the hard dive as well a one dithered down to 16bit. 

As expected I can see the lights of the peak bit meter in X-ISM dancing all the way down to the bottom of the scale for the 24 bit files (no indication of course if its only noise down there or where that noise floor may be). The lights of the peak bit meter appear solid from the 8th bit down for the 16 bit file, which is also what I would expect.

24 bit file-


16 bit file-


Switching to the Bitter plugin, I was immediately drawn to its display meter because of the seeming ability to read the 'effective dynamic range' of the file off the meter. But on closer inspection, something seemed odd.  The meter was reporting active bits down to around the 32bit range for my 24bit file, with the level tapering off at that point. 

Here's a screen shot-


Yet other 24 bit files show an abrupt cut-off of activity below the 24th bit, which is what I would expect to see-


Similarly, the 16 bit file shows and abrupt cut-off of activity below the 16th bit-


Any idea what is doing on with bitter appearing to indicate more than 24 bits of information in some of these 24 bit files?  The tapering lower bit-depth cutoff seems to fluctuate with level.  Here are two shots from different portions of the same file showing the variation with program level:

High level (clipping indicated)-


Low level-

[Gutbucket]

OK I may have it figured out.  Bitter shows the extraneous extra bit depth info when I play FLAC compressed 24bit files directly in Foobar, but if I decompress the FLAC compressed file to WAV before playing it, Bitter shows no information below the bit depth of the information in the file (the expected behavior).  Whether this is an artifact of the FLAC decoding code in Foobar or what I have no idea.  I'd love to hear the thoughts of the group on this.

Here's the same file as before, played after decompression to WAV-

[Gutbucket]

Opened foobar on this machine again today and the Bitter meter was still selected.  I cued up an mp3 possibly sourced from a 24bit WAV and got the same odd bit-meter behavior, tapering from the below the 24th to about the 32nd

[live2496]

According to this Foobar2000 uses 64-bit float internally.
http://wiki.hydrogenaudio.org/index.php?title=Foobar2000:Playback_settings

In Preferences check your Playback->Output->Output Format/Post Processing.

Set it to 24-bit and then tell us what it does.

Gordon

[live2496]

Any news on this?

[Gutbucket]

Loaded Bitter on my DAW laptop.  In Samplitude it's quite cool to move the location of the meter in the chain and see the bit depth change.  With bitter pre-fader it displays the 24bit native depth of the recorded file with a sharp cutoff below.  Moving the meter to a post-fader position displays more bits in use with the meter tapering down to the 36 or so bit, showing the calculation depth used by Samp internally. 

Bitter in Foobar on this machine always displays 24 bits regardless of the output bit-depth setting in preferences and the input file.  I suppose that's showing the internal depth where DSP plug-ins fit into the path and any calculation to change output depth happens after that. The behavior is different than on that other machine and there are several possibilities why that could be, possibly different FLAC and Foobar versions among others, but I only plan to use the bit-meter in Samp so I'm really not concerned about it in Foobar.

In any case it's quite interesting to change the output bit level setting in Foobar down to 8-bits and hear the severe aliasing of the truncation, then click on the 'output dithering (slow)' option, restart the app to hear the audible dither noise and those artifacts disappear.

[live2496]

That's a revealing plugin. Thanks for the ideas!

It should take a lot of the mystery out of what is happening inside the software.