That's why I am looking at this thread and finding out why should I not just use the filter that goes down fastest/earliest for least amount of ultrasonic signal out of a DAC.
The filter's goal isn't to reduce HF, because most of it is part of the music. The key is to eliminate (or at least reduce) the portion of HF that is not part of the music -- that is noise. Theoretically, if the filter's stopband is Nyquist (22,050 for CD) then there won't be any HF noise.
Early / slow rolloff filters do have less passband amplitude ripple. But this is irrelevant because the amount of passband ripple with a well implemented fast rolloff filter is inaudible - hundredths of a dB. In that sense, a slow rolloff filter can be a cure that is worse than the disease. However, if the slow rolloff filter is applied at higher sampling rates, it can be perfectly flat up to 20 kHz which makes the slow rolloff harmless.
Take at look at the above filter graphs. 3 of them (Magenta, Blue and Brown) are fully attenuated by Nyquist. None of these will have any HF noise. Magenta is slow rolloff which is unnecessary. Blue or Brown is the one to pick - one probably is linear and the other minimum phase. Linear is the most "correct" yet some people prefer the sound of minimum phase. Most people can't hear any difference.
Now look at the two filters (red and teal) that don't fully attenuate until 24,100 Hz. These filters use a common engineering trick. They take advantage of the fact that each frequency's alias reflects around Nyquist to ensure that any and all HF noise it leaks through will be above 20,000 Hz. In other words, Nyquist (22,050) is the exact center of the filter's transition band (20,000 to 24,100). So there's no audible noise, and this gives these filters a transition band that is twice as wide, so they are cleaner and flatter in the passband. However, suppose this filter leaks energy at 23,000 Hz. The alias is 22,050 - (23,000 - 22,050) = 21,100 Hz. Both tones are inaudible. But their difference tone: 23,000 - 21,100 = 1,900 Hz is easily audible. So as these 2 supersonic tones pass through the analog preamp, power amp, and speaker/headphone, intermodulation distortion will create a 1,900 Hz tone. That's the bad news. The good news is that the filter attenuates these frequencies around 20 to 30 dB, and IM distortion is typically at least another 60 dB lower, so in most cases the 1,900 Hz distortion tone will be inaudible. And low enough in level that it won't hurt tweeters.
Overall, given the above filter choices, I'd use the red labeled "fast linear". It should be the most transparent in the passband and ensures that any HF noise is very low in level and above 20 kHz. But if you want to be super conservative about not allowing any HF noise at all, use the Blue (apodizing 5) or Brown (brickwall 7).