Technically meaningless AI word salad

[ampetrosillo]

https://techivation.com/blog/the-zero-c ... on-forever

There are no zero crossings in digital audio. Technically, the digital representation of the captured audio is a forever repeating spectrum from 0 to Nyquist, then reversed, then again, then reversed, etc. therefore there are zero crossings every infinitesimally small time interval.

[CinningBao]

Yeah, seems like it's trying to be too clever to be an April 1st thing. It's crazy how every dynamic management plugin has been doing it wrong for so long!

I'm surprised they don't rename themselves TechivAItion

[ampetrosillo]

But the thing is, it's not that if you only "squash" the wave after a zero crossing, then you are not generating any harmonics, or fewer harmonics. If you take a square wave and you squash only one of the two halves, what do you get? (Granted, it would only happen on one period of the wave, but for that period only, you are effectively distorting the square wave into a pulse wave with pulse cycle != 50%, plus some DC offset). The amount of harmonics generated doesn't necessarily correlate that well with the shape of the waveform, although discontinuities obviously generate lots of harmonics (but you can smooth those discontinuities out, of course).

[imrae]

I think it is saying their compressor introduces discontinuities in the first derivative of the signal, whereas conventional methods don't. That seems like a bad thing.

Also, the positions of these discontinuities will depend strongly on DC bias or low-frequency content, even if the detector sidechain is highpassed?

Really, what is special about zero versus some other reference voltage? Any of them can be "silent" if you hold the signal there.

[ampetrosillo]

It's not even that. According to their word salad, they are compressing after a zero crossing, full stop (as if zero were special, as you said earlier). That will introduce discontinuities even in the absolute signal.

[Dirk Diggler]

That would be completely wrong.
Every digital wave file will have a zero crossing for every cycle, end of story.

AI thinks by adding more information it will feel more believable. LOL

[Andreya_Autumn]

Well... If you'll pardon the pedantry, "there are no zero crossings in digital audio" is not the reason why that text is objectionable. Sure, the sample data itself only has a value at the samples themselves indeed. But that doesn't mean it's 0 in between, it has *no* value in between. And reasoning about what the reconstructed signal will be works just fine, a positive value followed by a negative one *does* mean a zero-crossing happens in between them (at some exact inter-sample location, perhaps even more than one, blablabla). And there *are* situations where that is useful to observe. Looping audio for example.

But all that aside, that techivation post is indeed really bad. Like, they need to market their stuff, sure I get that, but this is directly misleading/muddying the waters.

[ampetrosillo]

Well... If you'll pardon the pedantry, "there are no zero crossings in digital audio" is not the reason why that text is objectionable. Sure, the sample data itself only has a value at the samples themselves indeed. But that doesn't mean it's 0 in between, it has *no* value in between. And reasoning about what the reconstructed signal will be works just fine, a positive value followed by a negative one *does* mean a zero-crossing happens in between them (at some exact inter-sample location, perhaps even more than one, blablabla). And there *are* situations where that is useful to observe. Looping audio for example.

But all that aside, that techivation post is indeed really bad. Like, they need to market their stuff, sure I get that, but this is directly misleading/muddying the waters.

As I said, due to the nature of sampled digital audio (a train of impulses multiplied by the analog signal) there are actually infinite zero crossings for any time interval. When looping, IME, it's not zero crossings that make the loops sound better but it's actually the smoothness of the looping.

[DaveClark]

https://techivation.com/blog/the-zero-c ... on-forever

There are no zero crossings in digital audio. Technically, the digital representation of the captured audio is a forever repeating spectrum from 0 to Nyquist, then reversed, then again, then reversed, etc. therefore there are zero crossings every infinitesimally small time interval.

There is so much wrong with this description that I asked Google AI to write something in response to prevent wasting time:

"The statement is fundamentally flawed because it confuses frequency-domain sampling (DFT/FFT) with time-domain representation and misunderstands the nature of digital samples. Digital audio consists of discrete, finite amplitude samples taken at fixed intervals, not continuous, 'infinitesimally small' signal points with constant, infinite zero crossings.
Key Errors in the Statement:

Confusion of Domain: The 'repeating spectrum from 0 to Nyquist, then reversed' (often called 'mirroring' or 'folding') describes how the Discrete Fourier Transform (DFT) views a signal's frequency content. It is not how the captured waveform is stored or represented in the time domain.
'Infinitesimally Small Time Interval': This is incorrect. Digital audio is discrete (sampled), meaning there is a defined, finite time between samples ( 1 / Sample-Rate). There is no information between samples.
'Zero Crossings': Because digital signals are discrete samples, the wave rarely crosses exactly zero between samples; it jumps over it. Therefore, you do not have constant, infinite zero crossings, but rather specific sample points that might be near or at zero.
Misinterpreting 'Digital Representation': The statement tries to apply continuous analog theory (infinitesimally small) to a discrete, finite digital system.

In summary, the statement inaccurately applies spectral analysis theory to the time-domain sampling process."

[ampetrosillo]

https://techivation.com/blog/the-zero-c ... on-forever

There are no zero crossings in digital audio. Technically, the digital representation of the captured audio is a forever repeating spectrum from 0 to Nyquist, then reversed, then again, then reversed, etc. therefore there are zero crossings every infinitesimally small time interval.

There is so much wrong with this description that I asked Google AI to write something in response to prevent wasting time:

"The statement is fundamentally flawed because it confuses frequency-domain sampling (DFT/FFT) with time-domain representation and misunderstands the nature of digital samples. Digital audio consists of discrete, finite amplitude samples taken at fixed intervals, not continuous, 'infinitesimally small' signal points with constant, infinite zero crossings.
Key Errors in the Statement:

Confusion of Domain: The 'repeating spectrum from 0 to Nyquist, then reversed' (often called 'mirroring' or 'folding') describes how the Discrete Fourier Transform (DFT) views a signal's frequency content. It is not how the captured waveform is stored or represented in the time domain.
'Infinitesimally Small Time Interval': This is incorrect. Digital audio is discrete (sampled), meaning there is a defined, finite time between samples ( 1 / Sample-Rate). There is no information between samples.
'Zero Crossings': Because digital signals are discrete samples, the wave rarely crosses exactly zero between samples; it jumps over it. Therefore, you do not have constant, infinite zero crossings, but rather specific sample points that might be near or at zero.
Misinterpreting 'Digital Representation': The statement tries to apply continuous analog theory (infinitesimally small) to a discrete, finite digital system.

In summary, the statement inaccurately applies spectral analysis theory to the time-domain sampling process."

Fortunately for me, and unfortunately for you, Gemini doesn't really "understand" shit. And neither do you. Digital audio is only a representation (and a way of storing and processing such a representation). What it actually means is something else entirely.

[imrae]

Yeah, I disagree with ampetrosillo's claim but the fact they made a reference to frequency-domain picture doesn't mean they are confused enough to believe the signal is represented in Fourier space.

It's a disingenuous take from the LLM. I'd guess it was trying too hard to satisfy an instruction to identify flaws, so took a bad-faith interpretation as early as possible. But we can't ask what it was thinking. (Well, we can ask - but not trust the answer!)

[zerocrossing]

There are no zero crossings in digital audio.

Shut up! Shut up! Shut up! SHUT UP!

[Andreya_Autumn]

As I said, due to the nature of sampled digital audio (a train of impulses multiplied by the analog signal) there are actually infinite zero crossings for any time interval.

Sorry, I agree with the rest of your take but this just isn't true afaict. The sample train itself is completely valueless between samples. There is no signal there that can cross zero. There is also not zero. There is nothing.

[ampetrosillo]

As I said, due to the nature of sampled digital audio (a train of impulses multiplied by the analog signal) there are actually infinite zero crossings for any time interval.

Sorry, I agree with the rest of your take but this just isn't true afaict. The sample train itself is completely valueless between samples. There is no signal there that can cross zero. There is also not zero. There is nothing.

If there were "nothing" resampling wouldn't be possible. It is rather that there is nothing in the file, in the CPU, between two samples. But at a certain point, those integers or floating point values will have to be converted into a signal. If there were nothing, the reconstruction process would be basically invention. Instead the reconstruction filter (the ideal one at least) is simply a brickwall lowpass.

[cron]

Always funny to see AI reproduce common misconceptions as fact. Well, funny in an 'inevitable because of how LLMs work' kind of way. Say it with me kids: "audible clicks are caused by instantaneous changes to a waveform's angle, regardless of the instantaneous amplitude at which this change occurs"

I appreciate the AI might not be describing it particularly well, but if it works like the AI suggests (only changing gain at zero crossings) it sounds like they've built a zipper noise machine.