[sdiy] DACs on-board 12 vs external 16 vs 24 bit

cheater00 . cheater00 at gmail.com
Wed Nov 6 12:03:49 CET 2013

On Wed, Nov 6, 2013 at 11:44 AM, Tom Wiltshire <tom at electricdruid.net> wrote:
> Damian's answer of "do an experiment and see" is an excellent suggestion - take some good quality 24-bit audio and bitcrush it and see when you can tell the difference. But you asked the question in a general way, so I'll give you a general answer.
> The average human isn't going to hear much, if any, difference between 12, 16 and 24 bit DACs if the sample rate is reasonably high.

I would only be happy to make such a statement in the case of an
oscillator signal, which due to its very correlated power is going to
do funny things to our auditory system. On the other hand, a mono
sample of an acoustic instrument is going to sound bad at 12 bit, and
will likely start working better at 14 bit. That's when it's sitting
in the mix. Listening to a lone sample on a high fidelity system
you're going to need more than that.

For a full mix, 20 bit is necessary and the differences can be noticed
by a normal listener and pinpointed by a trained listener.

> I've experimented with audio at all of these bit widths (although never an A/B comparison, which would be important) and the quality of all of them is fine, and depends much more on other factors. Aliasing, for example. If you write some nice oscillator code that avoids aliasing and run it at a decent sample rate out through a 12-bit DAC, you'll get a more "high quality" result than if you do something ugly and play it back at 16-bit, complete with high-resolution aliased artefacts.
> So my experience has been that perceived quality is affected more by the algorithm than the DAC resolution, at least from 12-bits upwards.

Yes, non-linearities are perfectly acceptable if compared to aliasing,
because non-linearities create correlated artifacts whereas aliasing
doesn't. Except the bit crunch will alias too, so it's a mix of both.
A little bit of non-correlated artifacts and a lot of correlated

On Wed, Nov 6, 2013 at 11:59 AM, Richie Burnett
<rburnett at richieburnett.co.uk> wrote:
> save 24-bit for something like a synth that has to output 32 voices, each made of 4 oscillators all mixed together, with envelopes etc. The extra quantization levels are really useful here!

I seriously doubt 24-bit D-A is within the reach of any synth
designer. Or even 22-bit. Not even Yamaha would bother.


More information about the Synth-diy mailing list