I said that.

So regards "doing it wrong" I have been dithering prior to making an mp3...
I'm lucky it sounded fine, I suppose!

 

Excuse me, how do you know that 44.1k is needed for streaming? I've only heard of LUFS, and that even 16 bit streaming is required?
Do I have to downgrade my material to 44.1k? What for?

 

The units for measuring something are nothing to do with resolution.

 

yes, that is the part you failed to understand. math is abstract, physics is reality. it is two different things.

the question was "what is the frequency spectrum (or amplitude spectrum for that matter) of a digital signal", and your answer is "but that does not exist in reality".

so you are not answering the question.

a digital signal is not physics. it never exists in reality.

which does not change a thing that every professor who teaches math or DSP agrees upon a convention to talk about "negative frequencies" and use the -sign to display them.

no idea what you are talking about. are you?

negative frequencies occur in complex exponentials.

calling them negative frequencies and using the - sing is a convention which everyone agrees on. you would not even be able to explain SSB without them.

btw, there are even physicians which say that the origin of cosine waves are complex waves and not the other way round. but that is a rather philosophical question and has nothing to do with frequency spectrum.

that is the next thing you dont want to understand, that tones and spectrum are two different things.

there is lot you can represent in digital signals which can not be converted with a DA interface. this is not an argument, this is a helpless detour.

you can also not record a guitar with a max level of 1.5, yet you can represent that in float i.e. create such a signal content programmatically.

now you are argueing against your own nonsense and claim i would said that. classic.

that is the third thing you dont understand.

the question was which spectrum can be covered by a signal at a signal rate, not how to solve problems.

yes, exactly. and not 1Hz as proposed.

an interesting point btw, because in math the smallest number above 0, or to be exact, above 1/0 is still 1/inf. or to be exact, at least the difference is hard to imagine.^^

yet in practice there is a limit for frequencies in digital audio, which you correctly called ">0", and which is based on resolution and rate.

for example in a 32 bit 44 khz signal circuit you cant have a phase accumulator playing a sample (for) longer than 13 hours, so it is impossible to represent something which has a frequency of less than 0.00002136 Hz in practice.

that unfortunately means that you can not produce a quare shaped CV LFO for your eurorack synth with a speed of "1 day" for your next 300 days long ambient piece. you would have to use 64 bit audio for that.

a frequency spectrum is a range of numbers, for example "5-800".

it seems like you like the word "oscillation", that is fine, but that doesnt change a thing about noise having a spectrum, too.

for both, the amplitude spectrum and the phase spectrum, it is fundamental to understand what "positive" and "negative" means when it occurs, and to distinguish between the two and how to deal with them, because half the signal processes in music software requires that kind of stuff.

sometimes was there already, sometimes you did this yourself by mirroring frequencies intentionally, and sometimes only a little bit of negative spectrum content occurs because of errors.

it is a general misconception that 2-sided spectra would only occur in DFT; it occurs in every allpass or poles and zeros filter.

 

but would it be correct to only look at a part?

to retrieve frequency or power?

0 is a bitch, but not as brainfucking as -inf and all that shit.

 

For instance Spotify Accepts stereo FLAC or WAV. All files are converted to WAV (44.1 kHz) and transcoded into one of the following:
Ogg/Vorbis (96, 160 or 320 kbps) – AAC (128 or 256 kbps) – HE-AACv2 (24 kbps)

 

That's why it's called "lossless"…

 

Which is why I said: "negative frequencies is a purely mathematical artifact to maintain symmetry".

Which question?

Absolutely.

Negative frequencies are a symmetrical mirror image of positive frequencies. And they are not below 0Hz, because below 0Hz there is nothing. There is no -0Hz, because 0Hz is a state of rest, absolute silence! Negative frequencies are, as the SR describes it perfectly, above Nyquist. It is not called -22.05-22.05kHz, but 44.1kHz...
The sample rate is N - or the range between F0 (DC) and N if you will. Our useful spectrum is in the range between DC and N/2. So the positive frequencies are between >0Hz and ≤22.05kHz (or if we take it very precisely: ≩0Hz and ≨22.05kHz). Above that, the negative frequencies start in a symmetrical reflection, while the cycle rate of the negative frequencies decreases away from Nyquist.... We have to take care of the useful spectrum, not an imaginary spectrum over which we have no influence, except that we have to make sure that no harmonics are generated from the useful spectrum into the side spectrum region and that we keep our subrange around ≈0Hz clean. And I stand by my opinion that the concept of negative frequencies is a mathematical artifact that will not help anyone here unless he/she wants to deal with DSP or in another scientific field. But that was not the question of the OP, so why overcomplicate things unnecessarily? Negative frequencies are imaginary, they do not exist! It is nothing more than the second half of the amplitude of the positive spectrum, which is added to the positive at the end of each process.

Something I never claimed. I said: "a frequency spectrum consists of amplitude and phase angle of many single frequencies". In other words, the frequency spectrum is a function that describes a range of frequencies, their amplitude and phase angle, and, in the case of the decay spectrum, their course on the time axis. No matter what kind of signal spectrum we are talking about, it consists of sine waves. And we need sine waves to represent it.

1. Not really, since these two statements each refer to a different context. Plus I have corrected the missing reference with >0Hz, remember?
2. And now you accuse me of framing you for a statement you didn't make, while using that statement to frame me for a statement I didn't make. Funny!

--->

Don't tell me what I understand or don't understand. I still know best! I am always interested in educating myself more deeply. But now I ask you, do you really think that throwing complex functions into this space without any context-based additional information is a good idea? In context of such a simple question? I don't think so. You are assuming that everyone here is at least familiar with Euler's formula, which is definitely not the case.

If I'm not mistaken, you said it was 0Hz. 0Hz is absolute silence. Mathematically, it is correct, because the absence of amplitude can only be measured by including F0. But F0 is not a signal with amplitude and is not included in multiplications to calculate a plot. When was the last time you said, "Man, that sub at 0.5Hz is really gnarly," against the background that most mastering guys put the filter on at ~10Hz at the latest anyway?

If you say so...

If this were my native language, I would certainly express myself more eloquently. Instead, it takes me ages and countless corrections to write something that can be understood.
BTW: Oscillation is precise and noise is nothing but oscillation, as are all functions that have amplitude. I don't think anyone is disputing that these functions can have a limited frequency spectrum. I thought we were talking about the useful spectrum of the sampling rate, not an individual signal sound spectrum or the imaginary side spectrum.

 

Last edited: Sep 13, 2021

Not really, you won't hear the effects of dither on/off in most instances, as it's 60-90dB below the music . I recommend leaving it OFF for most situations. See also http://ethanwiner.com/dither.html

ofc, 96k = 2x CPU since samples/sec over 48k. Not including SRC conversions.

 

Here's an oxymoron... I was tempted to say oxmoron.... Lo-Fi is huge in a lot of modern releases.
It was originally meant as the opposite of Hi-Fi, meaning in opposition - low fidelity.
So based on the true meaning, anyone doing Lo-Fi, it should not matter, because the quality is in theory (by label) crap to start with.

 

Great insight about Aliasing..

 

Here's the truth in English and German:
Bit rate in kilobits per second (kbps) - Bitrate in Kilobit pro Sekunde (kbps)

 

I'm an absolute fan of FLAC and OGG Vorbis. The problem with both formats is that they are not widely supported as many people - including me - would like it...

 

Hello... If the project is already at a sample rate of 24 bit 48000 Hz, are there any side effects (eg artifacts or noise) when I bounce it to a sample rate of 44100 Hz using a dither on mastering?

Thanks in advance gentlemen

 

well, you add didn't you

 

-

 

Last edited: Jun 4, 2024

I mean unwanted sound artifacts that are clearly audible in the conversion output.

Ahh yes, I just remembered

Does it also work fine in reverse?
For example from 44100 to 48000?

 

dither only required when going down on the bit rate, and only on a mastering stage, if for some reason you need to render at 16bit then dither is required, otherwise just don't.

personally i mix at 44.1 and render at 96 for best quality possible.

 

this is what I meant earlier. Mostly inaudible though.

 

If there are already samples or recordings in 44.1kHz in the project, this is not a good idea because it will affect the samples' timing. Try upsampling a DIRAC and you'll see.