16 Bit vs 24 bit tested

[Airwaves]

Responding to Sweetwater In Sync Aug 16/01

For me, the 24 bit recording setting in Cubase 5/Win is audibly better than 16 bit.

I tested with an acoustic guitar, Shure KSM32 mic, 20 bit analog input Gina. Recorded two tracks: 1 at 16 bit and 1 at 24 bit. Same guitar part same tempo.

Result was that the high end sounded more natural at 24 bit. The 16 bit track had buzzy/fake sounding upper frequencies. When the 24 bit track was dithered back to 16 bit for CD, the original and dithered tracks sounded close enough for me.

This may only be a Cubase thing but all my new tracks are now fully 24 bit. Boy does it make the drive work harder!

[DAS]

Sorry, we weren't clear enough in inSync about the bit/sample rate thread. It is actually located above in the Question of the Week section.

http://www.sweetwater.com/forum/forumdisplay.php?s=&forumid=42


However it is also a great topic for this forum. Your experience interests me. Normally bit depth differences wouldn't have the effect you describe on high frequencies except at very low levels. Can you provide any more detail about what you heard, and and the recording levels involved?

[sreder]

It is overkill to record 24-bit audio files when the Gina has only 20-bit AD converters (I used to use a Layla with the same specs). Also, when recording at 16-bit, 4 bits are being truncated. Just something to consider with this test...

[sreder]

Just a clarification on the previous post, you can only choose 16-bit or 24-bit in Cubase (if this is still the case in version 5), so matching up with the sound card at 20-bit is not an option. Also, when I mentioned truncation, I was referring to how recording at 20-bit or 24-bit and then using a dithering algorithm (Waves IDR, Apogee UV22, etc.) prior to going down to 16-bits can result in better sound quality in comparison to just truncating bits without dithering. Of course, using a 24-bit file dithered and then converted to 16-bit would defeat the purpose of comparing audio quality when recording at 16-bit and 24-bit!

[Shawn Parr]

Actually there can be a benefit to recording at 24 bit if your convertors only do 20 bit. There is an unfortunate limit on dynamic range imposed by analog circuits preceding and following any digital convertor. Because of this the best specification I have seen to date is only 117dB of dynamic range. If you do the math, 20 bits has a theoretical limit of 120dB of dynamic range. Therefore if you record with the best convertors around you should capture with the same fidelity whether you use 20 bit or 24 bit.

There are many reasons that DAW's don't tend to have a 20 bit setting. One of which is that since computers have a tendancy to work with bytes (which is a set of 8 bits) it is much more efficient to deal with 24 bits (3 bytes, or words).

Another reason is that the internal bus resolution doesn't suffer from the limitations stated above. Therefore when processing most mix busses and plug-ins do their processing at 32, 48, or even 56 bit resolutions. They then typically dither back to either 32 or 24 bit depending on the plug-in and DAW to be routed/mixed around. This higher resolution can be utilized to maintain a higher quality signal until the signal needs to go through a convertor and into the analog domain.

:)

Shawn Parr

[sreder]

Shawn: Based on the facts you present, there wouldn't be a benefit to recording at 24-bit resolution with 20-bit converters, there would be no audible difference when recording at 20-bit or 24-bit, right? It would also waste about 1.2 megabytes of storage space per mono track per minute, which is why I called this overkill.

So why bother with 24-bit and/or 32-bit floating point files? I have heard a few arguments that higher sampling rates improve fidelity due to better representation of harmonic content (many inaudible ones that have relatives in the audible range), but no arguments for the higher bit depths other than to improve sound quality in digital signal processing (as you mentioned). Maybe digital audio companies have an inside deal with hard disk storage companies...

[Shawn Parr]

There is an audible difference between 20 bit and 16 bit. 20 bit is better than 16 bit from my personal listening tests. The only way to capture this to disc is with software that does either 32 bit or 24 bit. Other wise you are either truncating (BAD!!) or dithering down (adding noise & losing resolution later). I definitely think that going to 24 bit is worthwhile for a lot of music. For some it may not be, that is your decision to make.

Anymore when I am working with digital audio I use 24 bit for 2 reasons:

1. It sounds better up front, especially on acoustic instruments/voice

2. To preserve quality through processing stages (I do a lot of processing :) )

Storage anymore shouldn't be a concern. I remember when I was doing theatre for a living and just getting started having to scrape up $600 for a 1 gig drive for my Sound Designer II system . . . (I'm probably dating myself here) Now I could get increadibly huge drives for that kind of money.

[tubejay]

You can record at 24 bit in cubase and it will truncate. You'll end up with a 24 bit file that actually only uses 20 bits. Correct me if I'm wrong, but the resolution of 16 bit is roughly 65000 pictures and the resolution of 20 bit is roughly 1,000,000 picutres. Which makes a huge difference in clarity, openess and acuracy of high frequencies. It really is a huge difference and well worth it. 16 bit is total crap. It creates distortion in 15 khz to 20 khz range.

[recordingpro]

Just a note, I think that the improvement in sound quality from 16 to 20 is actually greater than the improvement from 20-24. I tracked albums for 2 years at 20 bit and loved it. Of course, I'm at 24 bit now, and loving it even more :)

[Shawn Parr]

Tubejay, you are correct in your math with 16 bit having a little over 65,000 values, and 20 bit having more than 1 million. The part that you may need to rethink is the pictures part.

You may be thinking of sampling rate, that is what determines how many samples are taken per second, or pictures if you will. The bit depth determines how much information is contained in that picture. Your frequencies are time dependant, so they are recreated as the pictures are flipped through, kind of like in a cartoon flip book.

Here is another way to think of it . . .

Sample rate is how fast the pages of the flip book are flipped, more pages per second and you can get faster motion (i.e. higher frequencies)

Bit depth is similar to how thick the pencil lead is, 16 bit is much thicker than 20 bit, so with 20 bit you can add shading to get more contrast in the pictures.

Now with both Sample Rate, and bit depth you can get better results by raising both. With current systems you get a much greater benifit from using 20-24 bits over 16 than you do by using 96khz over 48khz. I'm not saying that there are no benifits to 96K, just that you are more likely to notice the difference with the 20 bit. This is because you are adding 8x the amount of possible dynamic range to your signal (96dB - 120dB, every 6dB being double).

As for the sound card issues, if your card captures up to 20bit, but your software only does 24 or 16, by using the 24 setting all 20 bits will make it into your file as 24bit files with zeros added in the 4 bits not available. At that point the 24 or 32 bit internal path can use the available headroom while mixing and doing effects to maintain a higher fidelity.

:)

Shawn Parr

[blacko3788]

even though a 20 bit inputs are being used it's best to record higher bit.....I've give an example:
let's say in the analog rhelm you have a lesser quality mic; would you plug it into the same quality board and record on the same quality recorder? NO! So why not use the best quality bit factor for recording digital? it will just make the twenty bit sound more like twenty bit without loosing any more dynamic range.

[sreder]

I disagree. Recording a source that is only capable of populating 20 bits per sample to a 24-bit file has no benefit. It doesn't hurt anything, but it has no benefit.

I also saw a previous post that referred to 24 bits being equivalent to three bytes or words. The byte part is correct, i.e. 24-bits = 3 bytes, but personal computers haven't used 8-bit words in a long time. For example, the current generation of Intel processors use 32-bit words, with a "word" loosely defined as the amount of data that CPU processes at a time.

This is all very cool discussion!