Hacker News Reader

Comment by ReactiveJelly | Hacker News Reader

So if you had an 88k recording, you could run it through a well-known anti-aliasing filter to create a 44k recording that sounded the same?

So the only situation where 44k and 88k can sound wrong is if... the 44k file is different and wrong?

View on HN

If you think about this "aliasing" as in, what occurs in 3d graphics, then you can understand this. What these 3d fiters do is either remove infirmation with blur (FXAA) or use information that is not available in the image (MSAA and derivatives)

In audio recording, sampling at 88k would be like generating MSAA x2 image, so it can be displayed with higer fidelity, despite the outgut resolution being in lower 44k sampling rate.

Aliasing can actually occur in the capture stage if the setup isn't right. Think of moire patterns in a video with highly textured subjects. It can happen whenever a signal is sampled.

The mastering discussed higher up in the thread is going on ahead of time at the studio, not on your playback system. The whole mastering pipeline starts with some initial capture resolution from microphones/cameras. The studio processes these original raw captures into a combined form and prepares the distribution format, i.e. a planned audio/video stream resolution. The studio can use different resolutions during capture, processing, and final distribution.

Generally speaking, the highest rates would be easiest to work with and avoid perceptible artifacts. But practical tradeoffs are made to save cost whether in processing, transfer, or storage.

The point is that because of sampling, order of operations can matter. So having a 88k file -> apply an effect -> downsample to 44k, can sound different than having a 88k file -> downsample to 44k -> apply an effect.

This is an important point. The main reason that pro audio gear pushes bit depth and sample rate up to higher that 16/44.1 audio is because when you start doing the floating point math to mix and apply effects to audio you can end up with audible differences when multitrack recording. In this case (and I still think it’s optional for all but the most demanding recording of live performance) higher sample rates can help and to a lesser degree but depth can give you more dynamic range.

I give that long preamble to say once a record is done and mastered, having > 16/44.1khz is wasted bandwidth.

You can verify this by mixing to mono or splitting stereo and inverting the "after" and mixing them back into the "before".

If you get silence, they're perfectly identical.

Being pedantic here, but since it's on topic. This only applies for non-linear processing, which is most of what we want to do when mixing music. But not exclusively.

Where would “taking a 44k track and up-sampling it to 88k (audio DLSS?), applying effects, and then downsampling back to 44k” fall between those two points in the spectrum?

The downsampled 44k that went through a half rate filter might actually sound better, for that matter. The speakers won't try to reproduce the content above 22khz then.

Even in professional audio environments, monitor speakers' electronics are generally designed to filter out such high frequencies.

The "frequency response" spec listed on speakers will tell you what range they are designed to reproduce. Typically, it's approximately 20Hz to 20,000Hz to match human hearing, perhaps with a higher floor if the speaker is designed to be paired with a subwoofer. This range is usually a deliberate (and sensible) limitation imposed by the electronics, not necessarily the materials or the magnet design, etc.

Some speaker manufacturers will list abnormally low or high range numbers on the spec sheet in an attempt to attract customers who mistakenly believe a wider range means the speaker is better. But even those speakers have a steep roll off curve at the extreme ends of the range, so it barely makes any difference.

Even pro audio manufacturer don't seem to lowpass their loudspeakers. Compounded with aluminium dome tweeters having a breakup resonance around ~25 kHz, the problem /could/ exist; but since we don't hear above 20 kHz (and not much above 18~19 for most of us), the only possibility would some kind of mythical modulation I've never seen measured in real world design.

Example: https://www.soundandrecording.de/app/uploads/2020/10/8361-FR... from https://www.soundandrecording.de/equipment/genelec-8361a-3-w...

Reply on news.ycombinator.com