What is the purpose of the three lines in Har-Bal?

[har-bal]

The three traces are inexorably connected so that if you apply a filter to one that same filter is applied to all. To have a clear idea on what the difference is between the peak and the average spectrums requires a bit of an understanding on how the spectrum is measured in the first place.

The way Har-Bal does it is the traditional FFT spectrum estimation approach of taking windows (short time frames 16384 samples long) and calculating a periodogram for it.
A periodogram is just the magnitude squared of the discrete Fourier transform of the windowed data. Now the average spectrum comes about through averaging all periodograms together for the entire track. The peak spectrum comes from taking the peak spectrum magnitude from all periodograms for the entire track. The red trace is just the geometric mean between the two.

Why Har-Bal records both relates to how, in some music, problem resonances can stand out more in the peak spectrum than in the average one.
An example might be a particularly strong instrument solo in the middle of a track that has one or two really loud resonances of short duration. If loud enough it will stand out in the peak spectrum but if they don't last long then they wont be very visible in the average one.
Such a loud resonant peak might well be uncomfortable to listen to but if all you had to go by was the average trace then you might well find it difficult to figure out where the problem lies.

What we've found is that if you want the dynamics of the track to sound controlled then concentrate on the peak spectrum trace (yellow) and if you want to preserve the dynamics but just get the balance right use the average spectrum trace (green) and if you want a compromise use the geometric mean trace (red).

Paavo

[mflorio]

Paavo,

It seems to me that when peaks are reduced, the average and mean traces are correspondingly reduced by about the same amount. Why should this be ? I realize that all the traces need to be re-calculated, but it seems that if I'm reducing a peak (just one periodogram), the average should not be affected much at all (but the mean would). Yet my average (green) traces seem to be reduced by about the same magnitude ?

Mike

[HarBal]

Mike,

HarBal is a static linear filter. The frequency response does not change with time. Thus the filter will have the same effect on all traces.

A simple validation of this is to design your filter, apply the changes by writing the output to file and then opening it up with HarBal. Now compare the analysis of the written changes with your designed changes. They should largely line up except for the LF resolution issue.

Don't confuse what Harbal does with dynamic EQ. It definitely isn't that.

Regards,


Paavo.

[eyesore]

Can you elaborate on how the two spectra are calculated? I think I might follow you on the average spectrum, but I am definitely not sure what you mean for the peak spectrum.

Thanks.

[HarBal]

I suppose the easiest way to view it is as follows.

Consider you are looking at a 1/6 octave real time analyzer. If you make a movie out of the analyzer display as you play the track, then the average trace is the same as taking each bin/band value from each movie frame and averaging their values together to come up with an average bin value. The peak trace is the value of the highest bin/band level out of all of the movie frames.

That is basically what is going on though the exact mathematics is a bit different. It is a good analogy though.

Cheers,


Paavo.