So I have a theoretical question that's been bugging me for awhile.  Maybe for another thread, but I can move it later if anyone thinks it's worth exploring.  The general common consensus is that the digital domain gives us perfectly accurate audio reproduction.  Bits are bits and all that.  Yet, digital appliances like DACs seem to often employ either a specific selection or choice of digital filters.  So the question is, if digital audio reproduction is perfectly accurate, why do we need the existence of different filters and which filter represents the 'correct' one?

Filters in the DAC are needed to create a smooth transition from one sample voltage to the next sample voltage.
As DAC's must be capable of handling various bit-rates this 'smoothing' can not be done correctly with a fixed analog filter as that would need to change it's cut-off point depending on the sample frequency which is not practical/feasible in the analog plane but rather easy to do in the digital domain as these DAC's 'work' on much higher frequencies anyway. As there are a few ways to Rome and people like to have choices different filters are easily made and selectable.

But in essence you are right in assuming that even when the audio-data enters all DAC circuits in a similar way (otherwise it wouldn't be bit perfect) the analog waveform is NOT exactly the same when it comes to the frequencies that come near the nyquist frequency of the signal.

The general common consensus is that the digital domain gives us perfectly accurate audio reproduction.  Bits are bits and all that.  Yet, digital appliances like DACs seem to often employ either a specific selection or choice of digital filters.  So the question is, if digital audio reproduction is perfectly accurate, why do we need the existence of different filters and which filter represents the 'correct' one?

You need an anti-aliasing filter. That's basically a steep low pass filter that removes (in practice: attenuates) frequencies above Nyquist.

Now there are some parameters to this filter.
One is phase. If you go min phase you will have quite a bit of phase shift in the audible range, because the filter needs to be steep.
In most DACs you'll therefore find linear phase FIR (finite impulse response) filters. No matter how steep the filter, the phase shift will be linear / the group delay will be constant.

Then we have slope/steepness which depends on the order of the filter. A steeper filter does therefore not only increase computational cost, but also means more ringing in the time domain. This ringing is usually not in the audible range.
There are some audiophiles that tell you reducing the ringing will improve sound quality, but all it will do is cause a more aliasing - into the audible range. And I can tell you, aliasing sounds really bad.

The cutoff frequency has to be high enough to not attenuate audible frequencies, but low enough to achieve high enough attenuation at Nyquist.

There's also stopband rejection. This is designed to be in line with the performance of the DAC. No point in using a 180 dB filter if the noise floor of the DAC is at -100 dB.

If you want to play around with those parameters I can recommend SoX. There's a SoX resampler plugin for foobar2000 too.




Imagine cutoff frequency = Nyquist. Everything after the green vertical line would be mirrored (aliased) back to below Nyquist.
This would be a damn bad anti-aliasing filter.

^ I may have skipped something but I didn't see them mentioning that all files were already filtered at creation, either in the ADC or later using a resampler, so you have no influence on that. No matter what playback filter you choose, you cannot undo the initial filter.

If the signal is already properly bandlimited applying another filter doesn't do much. You can however degrade the signal by using an apodizing filter that doesn't remove frequencies above Nyquist, i.e. doesn't remove all aliased frequencies.