Engineers often throw around the advice: ‘mix with your ears, not with your eyes’.
This is easier said than done in practice. We are heavily influenced by what we see when we’re working. Today Parametric eq’s plugins (i.e. Fabfilter Pro-Q) are the most common tool of mixing. Most of these plugins dispaly a dancing spectrum of frequencies as we listen back to music and make our changes. With all that activity and motion, it’s hard not to notice that visual display and use it as a reference point.
But what happens when that connection fails? Have you ever had the experience where you know something in the low end is too loud but you dont know where it is?
As we’re making these changes, we’re watching
For a long time our former teacher George Massenburg (inventor of the first parametric EQ) refused to update his EQ plug-in to have a graphic interface like you see in
One of the reasons is that scopes can be deceiving. That’s because showing the frequencies of sound is a tricky business.
Knowing how to work with scopes is a very useful skill in engineering. It’s true that our eyes can decieve our ears
Background
When we record a sound, all we have to go on is the vibrations of air, it’s either vibrating one way or the other way, as shown in a ‘waveform’.
[img]
So if we were to ask you what are the frequencies on that wave, how would you answer?
Hmm… that’s a tough one. Frequency is the number of vibratory cycles per second, so we have to do a few things
a) figure out where the cycles repeat
b) figure out how many seconds it took to make the repeat
c) then find the number of cycles per 1 second (divide the 1 cycle by however many seconds it took)
Ok… not so bad…
but them how about for this one?
[ complex ]
Oof, that’s a bit tricky it’s hard to see any pattern to this. So when we put that into a spectral anyliser, how does it know what the frequencies are?
And here’s the really tricky bit — if the frequencies are always changing how do we keep track of them?
Well luckily someone actually figured this out. The text book answer is that frequency scopes use the Fast Fourier Transform (FFT). The Fourier transform was formulated by mathmatician Jospeh Fourier. The basic idea is that all soundwaves can be described as a summed interaction of various sinusoidal frequencies. The blend of these frequencies build up into more complex waveform, and this can also be reversed using an approximation. That approximation is called the Fourier transform.
But in some cases it can be useful to
how does the meter work?
what is the fourier transform?
Settings
Type - specifies analysis types
- RT Avg - real-time spectrum analysis, producing averaged spectrum over specified peried (’Avg Time’ parameter)
- Max - cumulative maximum power spectrum (’infinite peak hold’)
- Avg - cumulative average power spectrum
- ‘RT max’ - real-time maxium spectrum with spectrum fall-off (for better meximum estimate it is suggested to use a higher ‘overlap’ setting)
- RT Sigma - sigma spectrum which shows how dynamic spectral are - best use is together with RT average mode. that way you can compare dynamic areas with spectral average levels. Highly dynamic areas will display above the average
FFT Size / Block size
- larger FFT Size means the spectrum is refreshed less frequently increasing latency
- Larger allows for more accurate frequency resolution in the low frequency range, but decreases time coherence — affecting the high frequency resolution
- higher frequency info becomes over-averaged
- High block sizes is also more processing intensive, so may require more latency buffer
- therefore if measuring a low frequency sound or a fundamental, use a high block size + higher overlap
- block size applies across the spectrum, a higher sample rate requires a higher block size, otherwise it will be lower resolution
Overlap
- controls time overlap between adjacent FFT blocks … 80% means the block being calculated is overlapped with the previously calculated block by 80% in time
- increases spectrum refresh frequency, used with higher block sizes to prevent the display from blinking,
- also requires more CPU
Time averaging / Average time
- can be set according to how fast you need the chart to update
- in RT averaging or max estimation, Averate time is an RT value… the time in ms for the spectrum level to fall down by 20dB
what are the differences between different windows?
Frequency slope
SPAN 4.5 db / octave- can make it look more elevated in the high frequencies
- this is just a diagonal line across the meter, increasing left to right. Contemporary music very closely follows this tilt decreasing left to right, so this allows us to evaluate the spectrum more easily
Windowing function
- Hann window - (used in Voxengo plugins), usually the best for music analysis because it offers superior separation between adjacent frequencies (especially important in lows)
- Hi- res window is non-conventional (”Nuttall” squared) used for technical analysis - possible to measure the noise floor and detect various spectral arteifacts
- Blackman - slightly better dynamic range in comparison to the Hann but also has a better separation between frequencies in comparison to Hi-res window
what is the best use case for different purposes?
- for music and speach (FFT size of 2048 or 4096 and averaging set to Fast or Medium will do fine)
- for test signals (pink noise, constant frequency tones), larger FFT (8192 oor 16384) with slow or infinite averaging works best.)
- best settings for pitch - voice analysis
social media idea
- release with a story/reel of a visualizer moving with the cello playing “these are sound frequencies, ever wonder how this works works?”
Would we actually be able to explain fourier transform to a general audience??
what are the settings used in popular EQ / visualizer plugins? Does this make a difference?
- Ableton
- MDW eq
- Fab-filter