Built in limiter/mb comp is used not only on mac based systems. I heard similar thing happening on my cheap asus laptop as well but I couldn't find it located anywhere on win 8 system

Yeah I hear the limiter on my macbook pro's speakers pumping once in a while.

It's only when you're really cranking them, and I imagine it prevents a lot of idiot kids from blowing them out within a week.

Pc Laptops have something called level correction that you can uncheck. Its in the control panel and sound. Click on your speaker icon then properties, then advanced, then enhancements. Its in that list. Im going from memory here so dont quote me on the exact steps.

Its definitely in all pc laptops though as part of windows 7 at least and its on as a default. Its some sort of auto gain pushing limiter. Its terrible sounding but once its turned off your music sounds fine again. I dont think its multiband cause it pumps like crazy.

Did you build that straw man all yourself?

I hope you know that the delivery requirements (I mix post) call for a maximum ceiling in dBTp, not dbFS. Actually most of the time it's -1dBTp (as in the R128 spec).

Calling the measurement "true peak" or TP really isn't true, because in interpolating the signal to develop the measurement, you have altered what you are reading, and it is no longer a true representation of the what you are sampling - in the true sense of the word true.

Maybe what oldanalogueguy was arguing. In any given digital stream, there are NO intersample peaks, because that runs counter to the definition of a digital signal. There is NOTHING in between the samples - period. So-called intersample peaks are the result of modifying the signal, such as with reconstruction lowpass filtering, etc, and can NEVER be a true representation of what you are measuring.

If you run a low frequency sine up to full scale, it is going to read 0 dBFS and it wouldn't meet the delivery spec unless you lowered it 2 dB. Whether you straw men want to refer to it as TP or FS is semantics AFAIC.

This was normalized in Audiofinder. you can see the sample point is at zero but the post DAC representation will not be. that I presume would be a peak reconstructed from the DAC which is higher than the given sample. given you can't account for how any given device might respond to it, is likely why industries like a little safety margin.

This is completely wrong and I'm a surprised to read such an insane nonsense in an audio mastering forum 2013. Similar to a self proclaimed race-driver who doesn't know the difference between a combustion engine and electric engine. It's ridiculous.

Of course there is something between the samples. In a discrete sampled signal, the "between" the samples is not a straight line to the next sample, the between is completely UNDEFINED. And of course, it involves serious overshoots in both directions. The thing that defines the area between the samples is the filter inside your DAC (i.e. the band-limiting). This is exactly what Nyquist said, but it seems to be surprisingly difficult understand, especially for audio engineers!

Just watch this video ASAP, it carefully explains all the stuff you misunderstood in the last 15 years!

Fabien! That is total rubbish. THERE IS NOTHING IN BETWEEN THE SAMPLES - that is the basic limitation of a digital signal. Digital means discrete time and discrete amplitude. There certainly -was- something between the sampling instants and the quantizing steps in the original continuous time continuous amplitude signal, but you lose that information when you "digitize" it.

So, I can't listen to a digital signal and neither can anyone else, so when we attempt to reconstruct a continuous time continuous amplitude signal that we can use, we have to fabricate what was lost, using an interpolating algorithm to fill-in what we don't have. If you have enough samples at a high enough sampling frequency, simple interpolation makes a pretty good guess at filling in the gaps, but it still is a guess.

You probably don't remember Wadia with their "french-curve" interpolating algorithm that they were claiming was able to extract more information from the bit stream than common interpolation. Well, that is all BS because interpolation is not extracting anything from the bit stream, it is manufacturing information that is not present.

Seriously, read about the Nyquist theorem, or at least, watch the video. You are totally off the mark. The video shows these things clearly.

And in case you are interested, a perfect interpolation methods definitely exists, it's the so called sinc interpolation and is PERFECT for all band-limited signals. Nyquist PROVED it, now it's your turn to do the opposite. Most *real* engineers learn these basics in the two first semesters. It's explained here and well established:

Nyquist–Shannon sampling theorem - Wikipedia, the free encyclopedia
Sinc function - Wikipedia, the free encyclopedia

Stop fooling people with your naive assumptions.

Fabien, perhaps you should re-read Nyquist. He said that theoretically you could reproduce the original waveform if you are sampling at MORE than twice the highest signal frequency AND you have infinite resolution. In practice, there is no perfect brickwall filter AND infinite resolution doesn't exist either.

Interpolate means "to alter or corrupt by inserting new or foreign material". It's right in the dictionary. You can't extract non-existant information by interpolation, you can only synthesize it. That's why you get the "intersample peaks". The interpolation doesn't know that you mucked around with the waveform in the digital domain, it just blindly sits there connecting the dots.

Maybe you should practice what you preach:

My understanding as well
A.

tpad, this stuff is proven.

"If a function x(t) contains no frequencies higher than B hertz, it is completely determined by giving its ordinates at a series of points spaced 1/(2B) seconds apart."

You don't understand the theory of interpolation. Of course one knows what's missing, the key part here is the term "band-limited", "discrete" and the fact that any signal is a sum of sine waves. Please look up these terms, they aren't fancy marketing terms.

Now don't come with "it isn't perfect in reality". It's still several orders of magnitude more accurate than any alternative method mankind came up with. And it doesn't explain your strange assumption that the super simplified waveform displayed on your screen has anything to do with the true digitally reconstructed waveform. It doesn't, and it's a classic mistake uneducated people do. However, as an audio "engineer", you should know better and at least try to understand the mathematical foundation of your main medium. I put it right in-front of your nose, and you still deny these facts, I think it's time to come up with some theory/explanation to support your steep claims or silently do your homework.

Got $36 ?????


Digital Audio Explained: For The Audio Engineer: Nika Aldrich: 9781419600012: Amazon.com: Books


I miss Nika on various forums trying to edumicate the masses! Where'd ya go Nika???

I understand where the confusion comes from because in every day language we use the term digital signal to mean the audio but that is not really correct. It is (when used by someone that understands the subject) just short-hand to save time and words.

The digital stream is not the signal itself. It is only an encoding a.k.a. representation of the signal we are interested in. And as with any encoding mechanism, you need to decode it to get the actual signal back.

With a good implementation of the decoder (using sinc interpolation), the band limited signal that was sampled can be entirely reconstructed in a fully deterministic fashion. There is no guessing involved.

This is fundamental to understanding digital audio!

And to continue in the same vein, the term digital signal does not mean the actual bits of a digital stream. A digital signal is the physical representation of the data within a digital bit stream. For instance the electrical voltage pulses in a copper cable used to encode the actual AES bit stream in an AES connection which in itself is an encoding of the analogue waveform that was sampled which itself is only an analogue representation of the air pressure waves we call sound.

When someone points a finger at the moon, don't look at the finger, look at the moon!

To get back to True Peak values, they are based on the decoded signal. If you use a True Peak limiter set to -2dB TP (for ATSC) or -1 dB TP (for R128) you will not have any issues with QC when delivering post mixes. (At least not due to this). If you use an old fashioned limiter that only looks at sample values instead of the actual signal, then you can get into trouble.

Alistair

Huh, what reconstructed waveform? There is no "digitally reconstructed" waveform in my DAW. If the metering in my DAW is accurate and it indicates that the highest peak value is just reaching full scale, I can open the corresponding WAV or AIFF data file and look at the sample values and verify that they indeed are just reaching full scale value. Basically, just a bunch of discrete amplitude numbers at discrete sampling time intervals that correspond to what is appearing in the DAW. There's no interpolation and no intersample values or intersample peaks. So, I don't have a clue as to what you are referring to.

If I do subsequently feed said data file through my DAC, the reconstructed waveform isn't "digitally" anymore, its analog, and there are no "sample" values or "intersample peaks" for that matter, because it is a continuos time waveform. Or at least it is with my DAC.

As to your lectures on interpolation, actually I think it was Claude Shannon not Harry Nyquist that popularized most of what you are referring to, and I am quite familiar with the theory (including interpolation). If you want be argumentative for the sake of argumentation, you probably will have better luck with somebody else - somebody with a spare 36 bucks to blow on the book :-).

This is all absolutely correct!

But as you yourself correctly pointed out, we don't hear the digital sample values. We only hear the audio that comes out of our DACs. To be even more correct we don't hear that either. We only hear the sound pressure waves coming out of our speakers. The analogue voltages in audio cables are themselves merely a representation of what we want to hear. For brevity sake I will from now on only refer to this whole chain of abstractions as "audio".

When building a digital level meter (and many other digital audio processing tools) it is computationally much cheaper to just look at the sample values rather than reconstruct the actual audio for every meter in a DAW. The result is that most digital meters are in fact inaccurate audio level meters. They are not even really audio level meters at all! They are just sample level meters!

We can't really blame the DAW builders though. For most practical purposes the sample level meters are close enough for every day use. And these "short-cuts" allow us to save much resources and therefore allow us to do many more things with the limited computational resources that we have. Without these kind of short-cuts, we wouldn't have all the wonderful tools we have today. At least not at this price and with so much power.

On the other hand, this saving of computer resources has brought us to the unfortunate situation that most people don't understand digital audio and confuse the sample values for the signal itself.

Indeed! But those peaks in the signal that was encoded in your DAW were at a time position that did not correspond directly to any sampling point. They were at a position that fell between two sample points. But it was still the same signal! It just happened to be in encoded form. The fact that your DAW took short-cuts in representing the signal (usually through a "join-the-dots representation") is what stopped you from seeing the Inter Sample Peaks. But they were there all the time!

Here is a screen-shot from a DAW that does show the reconstructed signal:



(Well at least it shows a better approximation than most other DAWs ).


The video Fabien linked is entirely free!

PS: I don't think Fabien or flatfinger are being argumentative for the sake of argumentation. This stuff represents a paradigm shift in understanding digital audio. I would even go so far as to say that people that haven't fully realised the implication simply do not understand digital audio.

Alistair

So if you have two continuous samples that clip, you necessarily have an intersample peak that will be "filled in" by the DAC?

This is what I recently wrote in another thread about a similar topic. I think it might help answer your question:

Not quite!

Exactly. It isn't about 0 dB FS samples. It is just that people notice it at the top because suddenly their floating point DAW shows values over 0 dB FS and they get mystified. Here is an example with maximum sample values of -1 dB FS:



(Bigger image here).

Note that this set of samples would create a peak in the reconstructed signal that is equivalent to nearly +4 dB FS or nearly 5 dB above the highest sample value. (See the dB scale on the right).

The important and fundamental thing to remember about PCM audio or MP3 or any other encoding format is that it is only a method to encode a signal. It is not the signal itself! The sample points you see in your DAW are just points on the signal's waveform. They are not the waveform! The only thing that counts ultimately is the actual signal after it gets reconstructed! That is what we listen to. That is what this is all about.

Just look again at the picture above. It shows the sample points and the actual signal that is encoded in those sample points. If you look you will see that the signal very often passes over the sample value. Sometimes only by a tiny little bit but still over the sample value. That is completely normal. That is just how sampling works. The sample points could be anywhere on the wave form. So obviously they will very often not be exactly on the peak of the waves. Strictly speaking, these are peaks between the samples or, Inter Sample Peaks.

In other words there are Inter Sample Peaks all over every single wave file. It is just that we do not normally refer to them as such unless they peak over 0 dB FS.


EDIT: I can't recommend the video that Fabien linked enough. It really does explain things in a very clear manner. I'll repost because it is so good:



I also have a thread in which I go into some of the details: Digital Audio and Sampling Rates


Alistair

Alistair, thanks, but what I was trying to explain (before the kibitzers showed up) is the apparent confusion or misunderstanding over the term "true peak". The real true peak value of the digital signal (bitstream) in your DAW is not necessarily the same "true peak" value that is being developed by an interpolation process.

Back in the 90s, Tektronix came out with a line of digital audio metering equipment with interpolation filters built in that were attempting to mimic the behavior of a DAC reconstruction filter. The marketing types were calling the resultant measurement "true peak". Unfortunate choice of terminology, because this has been misinterpreted by many to somehow suggest that what you are reading in your DAW is not an accurate reflection of the peak value of the signal in the digital domain (before it encounters interpolation). To the contrary.

If they had referred to the measurement as Interpolated Peak instead of this "True Peak" nonsense, and everyone understood what they really were measuring and seeing, then maybe these discussions wouldn't be necessary.

tpad, I give up. Have you ever heard of the Kruger Dunning effect? It can't hurt to be aware of it...

Good! Ever hear of the term "Officious" - it can't hurt to be aware of it.

Nice cheap shot, but he was right and you were confused & very rude. In the digital realm there is nothing between two digital samples.

You are right that analog peak levels after DAC are dependent on a DAC, not just the digital signal. However the term “true peak” — like it or not — is formally defined in BS.1770 standard as a peak value of the oversampled waveform (using a certain class of linear-phase oversampling filters), which allows it to be computed from a digital waveform. It may not be exactly the same as peak analog level after DAC, but in most cases it is a good approximation, a better one than the digital sample peak level.

So your technique caters only to other gear enthusiasts with a similar obsession with audio fidelity and quality?

You've got to consider the average listener these days, who is probably using a cheap iPod dock, laptop speakers, or whatever earbuds that came with their mp3 player.

That's their problem if they want to listen on junk!

Mastering a record for mediocrity is like building a 747 jet that will never be used in actual air service. Or a roller coaster with no hills. lol!

Umm, is that all there is? Just people listening on laptop speakers & earbuds or else mastering grade monitoring systems? Seems you've conveniently left out everyone else.

Is it really a good idea to make special concessions for people who want to overdrive playback on the crappiest systems? I mean at that point it's pretty crap anyways. Are those people even going to notice or care?

If you actually have a good mix the ISP's probably will be at insignificant amounts typically I think.

I don't think it really impacts things significantly one way or the other. It's always good to be mindful, keep tabs on what's going on, test different playback systems though.

Alexy - yes, I don't like a lot of the terminology being utilized. "True Peak" probably being one of the more egregious examples. I just call them the way I see them and then try and move on. Still wonder what DC was hinting at with Apple's multi-band compressor?

Why should it be a relationship between a 'good' mix and ISPs ? It sounds as silly as when people ask for what a spectrum analysis of a 'good' mix should look...
In my experience over 0 ISPs (the problematic ones) depend on how loud it was pushed and specially on the way loudness was achieved.

Hi tpad,

I understand where you are coming from but I am not sure I entirely agree. If the values given by the interpolation process are closer to the levels of the actual signal we are interested in (meaning the decoded signal) compared to a simple sample value reading, isn't that desirable? And isn't it then a truer representation of the peak level of the signal we are interested in? (Again, the sample values are not the signal. They are just the signal in encoded form).

Here is yet another analogy: If you received a poem in encrypted form, would you judge its beauty by looking at the encrypted text or would you decrypt it first?

Alistair