Why not always cut the 20-30 Hz range?

[svenneke]

Okay, for those who though I was dead serious when I wrote:

"0 dB/0? Are you serious?
Dividing by zero is not allowed in mathematics. You should know that."

I wasn't of course, it was a joke... .

I know that aciddose meant 0 dB/octave but even in that case it is not correct.

0 dB/octave means no level change at all. So that is a pretty useless filter.

0 Hz means no vibration, no movement, no real frequency.

A frequency can come close to zero and it will result in a very long period (or wavelenght if you take the speed thru the medium into account) in that case. But it cannot be zero, otherwise there is no period at all. So theoretical you can speak about 0 Hz, but in reality it is no longer a moving, cycling, alternating phenomenon anymore at that point, just a static value. I hope this will expain things a little bit more.

If anyone still has a question about DC offset or why I use highpass filtering
on almost all track, shoot.

Sven

[aciddose]

what the hell are you talking about? i wrote 1/inf db/o.

we agree fundamentally but i think you make a lot of assertions that you can't back up. regarding "dc offsets" "don't use a filter for dc offsets" which i think is nonsense, and "use a steep filter" which is ignorant of the specific context in which it's being applied.

you can't hand out generic advise like this without backing it up with a specific context, otherwise you're just giving people bad advise.

[rifftrax]

Hi,

I see all kind of information here, some right, some wrong.
Let me share some info with you.

Is that right? I'm all ears to hear what types of 'wrong' info you've detected.

First DC offset removing should not be done with an equaliser but with a DC removal plugin. If your DAW does not do that, which I doubt because most modern DAWs do, download audacity (windows, mac and linux compatible) and use the normalize function. This process allows you to remove the DC offset too. DC offset is a slightly level shift away from the zero crossing line towards the upper or lower side of a soundwave. An equaliser cannot change that shift because it is not a frequency, it's a shift in level away from zero. Only a specialised plugin can do that.

Sorry. You're wrong. SonEQ, Fruity Parametric EQ, Fruity Parametric EQ 2, BootEQmkII, TRacks Classic EQ, CheifGEQ, Overtone GEQ, and Phase EQ all are able to or cut DC as a default. DC offset removal does not need to be done by a "specialized plugin" anymore than a simple square wave needs to be created by a "specialized synthesizer".

I almost always use a high pass filter on any track. The reason is that low frequency content takes away most of the headroom of a track.

Really? Well, this definitely sounds very scientific.

The amplitude of a 50 Hz wave need to be much higher than the amplitude of let's say a 3000 Hz wave to be heard equally loud. Therefore woofers are build because they can move more air. A tweeter can hardly be seen moving forwards and backwards and yet we can hear them very well.

If this were actually true - people would need to hit the low notes on a piano harder than than high notes for piano solos to make sense (hell, even try it with sine waves in a basic synthesizer if you're worried about the effects of harmonics... viz. C1 and C4 on a keyboard at the same db level in any sequencer for a pure sine tone sound pretty damn well matched to me). In addition your understanding of voltage waveforms vs. what is required with a speaker transducer in terms of excursion/air displacement to achieve similar spls for varying frequencies is fundamentally wrong. A comparable amount of linear excursion (or air displacement) within a single speaker diaphram for different frequencies (say 50hz vs. 1000hz) definitely does not equate to the same sound pressure level when assessed using an unweighted measurement from a decibel meter.

If you look at the dynamic range a 16 bit file can hold, there is much less space for multiple soundwaves to use in the lower frequencies, speaking about volume, than in the mid and higher frequencies because the big low waves simply take to much space (amplitude value is high). If you reach the upper limit, you will get distortion because all other frequencies from your mix will add to that signal.

What in the world are you talking about? Honestly? This is COMPLETELY inaccurate information.

So cleaning up everything below what you need in each track, creates space for the rest to sum up. It's as simple as that. Look with a frequency analyzer at a recording of your voice speaking some words with P and B's in it. You will see the low content when those two letters are spoken. If you zoom in on the waveform in your DAW, you will find at those letters a few big long bumps in the signal where all the rest of the sound is modulated by. This is the impact of low frequencies.

No, this is simply the effect of any frequency mixed with any other frequency. Normally plosives as captured by a mic with no pop-filter are going to overload a microphone however... and this is a separate problem.

Now, imagine that this can happen in all live recordings with background rumble, resonances caused by guitar body's or drum sets to name a few and everything that produce low frequencies will mess up the bottom of your mix. It might look as if it's not there because you don't hear it or your speakers cannot reproduce it loud enough, but it is present in your data and it consumes energy and space. That energy is waisted and can cause mixes that do not sound as clean as they could be.
At the same time it prevents getting a decent loud mix and master. The cleaner each track is, the louder you can mix. (not that I am a loudness junk, au contraire mes amis)

Or maybe, filtering out frequencies created by hits on a drum kit, simply make it sound less like a real drum kit and more like a filtered/processed drum kit. Not that there is anything wrong with that, if you love the sound of tons of processing or your monitoring environment sucks to the point that you can't monitor accurately below 40hz (also, a separate problem).

So yes, always clean everything below 30 - 40 Hz and even higher if your track allows you to do. Your results will only sound better and cleaner.

SVen

Well, that is some kind of magic bullet. I'm glad you've got this so well figured out. OR maybe your advice is total bullshit based on a very distinct set of misunderstandings of audio engineering concepts.

[svenneke]

what the hell are you talking about? i wrote 1/inf db/o.

we agree fundamentally but i think you make a lot of assertions that you can't back up. regarding "dc offsets" "don't use a filter for dc offsets" which i think is nonsense, and "use a steep filter" which is ignorant of the specific context in which it's being applied.

you can't hand out generic advise like this without backing it up with a specific context, otherwise you're just giving people bad advise.

Hi aciddose,

can you explain then to me what the steps are before coming to a statement
1/inf db/o?
Maybe this is a language problem for me, english isn't my native language, but is it correct to read this as:
1 divided by infinte db per octave?
In that case can you give some links to info that explain this formula?
I am here to learn also so I like to be constructive.

I admit that you can use a high pass filter, I was wrong in writing this, sorry for this misunderstanding.

My point was that there is a tool that is created for that, so why should you use a extra filter.

About that steep filter I think I explained it clearly that I would use that to eliminate frequencies below 20 Hz with a stronger attenuation than with a gentle filter. Not that a gentle filter cannot be sufficient, but if there is a strong frequency at for example 15 Hz, you can eliminate or at least attenuate it better with a steeper filter @ 20 Hz than a gentle one @ 20 Hz, that's all I tried to explain.

Your advice to explain things better is noticed, I will try to do that to prevent any further misunderstandings about what we are talking.

Sven

[aciddose]

can you explain then to me what the steps are before coming to a statement
1/inf db/o?
Maybe this is a language problem for me, english isn't my native language, but is it correct to read this as:
1 divided by infinte db per octave?
In that case can you give some links to info that explain this formula?
I am here to learn also so I like to be constructive.

it means yes, 1/inf, db/o. so it would be the smallest possible slope of a filter. of course if you're using an ordinary integrator the minimum is 6db/o, so it by far satisfies the requirements to eliminate dc to a level of -inf. it's actually inf times greater slope than required

My point was that there is a tool that is created for that, so why should you use a extra filter.

i tried to explain that i think when people say "dc" they are rarely talking about a property that can be eliminated by adding or subtracting a constant offset.

of course you are aware that it is impossible for any offset to be constant, at most it can take up the complete "window" of a sample or period of time.

while it would be in some cases best to calculate the average value, then subtract this to eliminate a "true dc offset", it will rarely work because most samples will not contain an offset that is constant.

in cases where you are dealing with short samples, it's likely best to apply a dc removal (average and subtract), followed by a highpass filter. it can be worse though! imagine this - the beginning of the sample has zero offset, the middle and end have significant offset. now the offset correction will create an impulse at the beginning of the sample equal to the negative average of the complete sample, so it will introduce a "click". if you use only a highpass filter in this case no click will be introduced unless it already did exist.

in all other cases however such as longer recordings or complete tracks you must use a highpass filter unless you know specifically that there is a constant offset. since in complete tracks it's very common to have changing levels and various types of effects processing, it's extremely unlikely that the offset will be "true dc".

i put "true dc" in scare-quotes because like i said it is impossible for any value to really be constant for infinite time. what we have to do is assume this to be true, and then change our definition of "dc" to always refer to "pulsed dc" or "constant inside a window" as i explained i believe most people intend to mean when they say "dc".

[svenneke]

@ rifftrax

Thank you for your feedback.

See my previous post about using a highpass filter, it can be done and I was wrong, point taken and sorry for that.

Let me explain what I mean when I wrote that low frequency content takes away most of the headroom of a track. It is a little bit of science btw.

When you look at the waveform of a mix, the peaks or transients of the drums and other transient-rich material can easily be seen.

Now, let's suppose we compare the attack of a kick drum low end to the transient of a piano note played at C4.
We will only look at one period of the base frequency of both, at full intensity right after hitting the kick and the piano key. After that the natural decay starts to attenuate the sound pretty fast.

Suppose the low end of your kick drum has a frequency at about 60 Hz.
The fundamental frequency (also the loudest) of a C4 piano note is 261.626 Hz.

The time it takes for one whole cycle or period is 1/f (f=frequency)

So, 1/60Hz = 0.016666 seconds and 1/261.626Hz = 0.003822 seconds.

This means that the time it takes for one whole period for the kick drum's lowest frequency is 4.36 times longer than that of the C4 piano note's fundamental.
So therefore I think it's safe to conclude that during the kick drum there is less headroom for other frequencies because they sum up to that kick drum's frequency. The longer the wavelength, the longer it occupies the space in the mix.

Isn't this the reason why you can find hundreds of pages about sidechaining your kick drum? When the kick comes in it uses so much space that you must make room for it in a busy mix.

So this is what I was trying to expain, but didn't do it as clear as I should.

For the rest of your remarks, I cannot be here for a few days and have to leave now ASAP, but I'll be back to explain in more detail a few things.

Oh, one more thing, the reason why people do not need to hit the low notes on a piano harder than the high notes is because the hammers are adapted from the low/left side to the high/right side. There is a gradual difference in size and weight and spring construction to be able to give more energy to the thick strings on the left and less on the right.
http://www.pianofinders.com/educational/touchweight.htm for more information about a few things.

Bye

[svenneke]

Yes! aciddose, now we are talking about the same things

If I understand it correctly what your mean it is essentially this: create a filter, the slightest slope will do, and let this function go to the limit, almost zero. The result is a nearly infinite attenuation of frequencies near zero, right? That explains a lot to me and perhaps others. Thx for your insight.

I agree completely with the rest of your post. I think our misunderstanding came because I was writing about a theoretical perfect constant DC offset, without any fluctuations. I know in real life there will always be slight differences and your way of dealing with that is absolutly right. However, I think if the fluctuations are so small that you hardly can measure them on a oscilloscope, in that case a DC removal plugin may do the job good enough.

Thx and bye
Sven

[svenneke]

Hi all,

I have some spare time so why not take some time to explain things a litlle bit more.

For those who were wondering what the hell I was talking about when I wrote:
0 dB/octave instead of 1/inf dB/octave?

I quote from a very interesting site:

"In math, when you hear people say things like "1 over infinity is
zero" what they are usually referring to is something called a limit.
They are just using a kind of shorthand, however. They do NOT mean
that 1 can actually be divided by infinity. Instead, they mean that,
if you divide 1 by successively higher numbers, the result becomes
closer and closer to 0. If I divide 1 by a very large number, like a
billion, then I get one-billionth, which is a VERY small number, but
it isn't 0. Since there is no largest number, I can always divide 1 by
a bigger number. But that will just produce an even smaller number,
right? It will NEVER produce 0, no matter how high I go. But since the
answer to the division is getting closer to and closer to 0, we say
that "the limit of the expression is zero." But we have still not
divided anything by infinity, since that isn't a number."

So far the quote. For those who are interested in reading the article click:
http://mathforum.org/library/drmath/view/62486.html

This is why I substituted (as a joke) 1/inf by the number zero.
What I wanted to see been written down instead was what aciddose answered after that:
"the smallest possible slope of a filter". Because in reality, at least in math and in real circuits, you will always use a real number for that slope, even how small it may be, but...

... on the other hand, in computer programming the "special case" float values +inf (0x7f800000) and -inf (0xff800000) do exist according to the IEEE-754 standard that deals with floating point operations in computational math.
See: http://www.cprogramming.com/tutorial/fl ... ation.html

So my question for aciddose is, what would you use to write the code for a plugin for such a filter? (I know you are developping decent plugins, see http://xhip.presetexchange.com)
Does using the +inf "special case" float has any advantages over using a real number value or is it simply impossible to use it for programming that way and is it only implemented in this IEEE-754 standard as a result of a calculation? I am interested to know that. Maybe you can explain this or you have already experience with that?
For those who think this is off topic, we are talking about writing a high pass filter so there can be some room for it, I hope.

Thank you
Sven

[aciddose]

the reason i said 1/inf was only to point out that dc is also an impossible such number. that's the reason i said "there is no such thing as dc". just as there is really no such thing as 1/inf. it's only intended to be used as an abstract when thinking about these situations.

in floating point it's used as a signal or flag. it doesn't in most cases have real applications because any computation done with inf will result in either another 'inf', or the error flag 'nan'.

the usefulness of that is it doesn't require constant checking of a special "flags" variable. instead, you just check the result of your computation directly to find if it has resulted in inf, nan, or other conditions. this means you can do many computations and defer checking until the end of that series. if a special flag variable were used and over-written by every operation, you couldn't defer such a check.

the filters i do use are the same as used in analog audio circuitry. a single capacitor and resistor which forms a 6db/o highpass filter will typically eliminate most issues.

when you find an issue that is not solved with such a simple filter, it tends to require working with that specific situation to find a solution that does work. there is usually no "fits all" solution.

[svenneke]

Thx aciddose for giving such usefull information.
I'm not a programmer myself but I do understand "a few" things

Sven

[svenneke]

Hi again,

To answer one of the remarks of a certain poster I like to take things a little further.
First I want to put my words into the right context about why I wrote that I almost always use a highpass filter on every track.
Of course I was talking about real live or studio recordings and not when people are using sampled instruments that are as clean as they can be most of the time. So let it be clear that you don't need to use such filter all the time, it highly depends on the source you use.

Now back on topic.
Plosives... What are they and why are they able to ruin your mix in the low end?

If we speak or sing words we are exhaling air, otherwise there could be no production of sound (vibrations) in the vocal cords. That airflow is relatively constant when the level of the sound remains stable. A bigger airflow will result in a louder sound an vice versa. The tension of the vocal cords determine the pitch of the sound whereby a higher tension also produces a higher frequency and vice versa. Certain letters are called stops or plosives because during these sounds we block the active air flow completely. By doing that we actually are creating a little bit of overpressure inside our mouth during a fraction of a second. Next we release the air right at the moment we open our lips. This happens within a timespan of about 20 milliseconds, depending on the type of plosive.

Bilabial plosives (p and b) requires bringing the two lips together. Dental or alveolar plosives (t and d) happen when the tongue closes against the upper teeth or the skin covering the roots (alveoli) of the teeth. Velar plosives (k and g) occur because the tongue closes against the soft part of the palate or velum.

There is also a distinction between voiceless plosives (p, t, k) and voiced plosives (b, d, g). Try this for yourself. Inhale and hold your breath. Now try to say p, t and k one after the other as long as possible. You will notice that this is not very hard to do. In fact you can do this as long as you can hold your breath because you only need a very small amount of airflow doing that and you don't use your vocal cords.
Now try the same with the letters b, d and g and you will immediately hear the difference. While making those plosives you also are making a sound that resembles to humming. You will need to inhale much faster now because you actually use a bigger amount of air coming from your lungs to produce that hum.

Back to our problem, regardless of the kind of plosive, what matters for mixing are those +/-20 milliseconds when the air is suddenly released. While doing that we create a small explosion of air pressure into the mic. That small explosion can easily be seen as a low frequency with a pretty high level if you zoom in at the waveform in your DAW or when you look at a frequency analyzer.
Maybe you wonder why we don't hear spoken words in the real world with those micro-explosions? Well, in fact we do if somebody is talking right into one of our ears at very close range. At normal distance when people speak to each other, that low frequency wave, with it's more omnidirectional dispersion, has enough space to disperse into the surrounding air and loose a bigger part of it's energy than the more directional higher frequencies. So we do hear those plosives in normal speaking conditions, but with a certain level ratio between high and low frequencies.

Unfortunately when recording a singer, he or she is also at close range to the microphone. Additionaly cardioid an figure-of-eight mics exhibit a proximity bass boost at closer ranges, whilst other types of microphones don't, and this way those type of mics will exaggerate this low frequency further the closer the singer comes. Singers with a good understanding of that effect can use it to shape the tone of their sound. Singers without experience often move constantly towards or away from the mic, not realising that this change the volume very much, but at the same time the shape of the tone. The use of a pop screen is adviced in that case to maintain a stable distance and so a stable level and toneshape. Some mics also have built-in highpass filters but I would prefer to use a better filter in your DAW on the recorded track.

Now the way I like to deal with those plosives is this. I do this in an audio editor and always before starting to mix in my DAW.

- Open the track and zoom in to the area where the plosive occurs and look for a sudden big bump in the waveform that decays pretty fast and that represents the low frequency plosive.
- Make a cut at the start and at the end of that bump and apply crossfades at both sides, some programmes will do this for you. Use a short crossfade at the start and a longer one where the low frequency starts to attenuate, in fact as long as it remains visible as a bump in the waveform.
- If your programme allows you to insert effects on a per clip base, insert a good quality equaliser into the effect slot of that clip between both cuts.
- Choose a highpass filter and set the frequency all the way down. Use a programme that recalculates the screen in real time so you can easily see the change in the resulting waveform.
- Now zoom out to a level where you can see a few words in the track, select a region long enough and activate your loop function.
- Raise the frequency of the highpass filter in your equaliser until you hear a natural sounding relationship between the whole word and the plosive. Don't eliminate the plosive completely because it will loose the information needed to be recognised as that letter.

This way you have total control over the length and the exact position of the plosive because many audio editers allow you to change the cuts and crossfades afterwards as much as you like and need.
But you have also completely control over the amount off attenuation you want to apply to that plosive only, without changing the level of the non-plosive frequencies present at that time.

Using a multiband compressor, as often is adviced, can of course attenuate the same region without altering the rest of the content of the sound. But this technique provides not such an easy visual control over the speed and position at which the crossfades allow the filter to start to attenuate and how it crossfades gently into the following part. You can even choose the shape of the crossfade that fits best for the beginning and the end of the clip. This technique does that in a very visual way, without any side effect. (perhaps a little bit of phase shift will be introduced, depending on your choise of filter, but with smooth crossfades and such a short time of about 20 miliseconds I doubt anybody can detect that). What I like about it is that you can see immediately that the bump locally starts to disappear as soon as you raise the frequency upwards.

This is also the right moment to clean up all unwanted low end and other disturbing content before rendering the track to a new file that can be used to start mixing. In many cases I also like to use spectral cleaning before rendering for everything that needs multiband compression otherwise. Simply because spectral editing is way more flexible than using fixed bands in a multibandcompressor.
But maybe that is for my next post if people are interested.

Have fun
Sven