[LAU] Text-based sound visualisation?

Fons Adriaensen fons at linuxaudio.org
Wed Jun 22 10:30:43 UTC 2011

On Wed, Jun 22, 2011 at 11:33:41AM +0200, Julien Claassen wrote:

>   One can listen to the bands using simple bandpass filters - probably a 
> bit broader than the original band - to see what is going on. So I get 
> the report 410Hz is very loud, I listen and see: Ah, it's indeed the 
> snare. then I can use EQ'ing on the snare, or more likely adjust it.

Why would you adjust the snare if it sounds OK ? Who cares that it
causes a peak at 410 Hz ? Unless you are mixing pinkish noise there
will be spectral peaks. They are supposed to be there :-)

And if the snare - or any other track - doesn't sound OK, then a
spectral peak display is not going to tell you why if you can't
find out without that. Forget the idea that a spectrum analyser
is some magic instrument that can be used as a substitute for
professional knowledge, experience, or your ears. 

Now if the point of the whole exercise is to reduce peak levels
while preserving loudness, and if it's the snare that is causing
the peak levels, that will be pretty obvious without a spectrum
display. Even if the only 'visual' info you have is something like
jack_meter. Changing the EQ a bit could be part of the solution,
as could be limiting, or just lowering the level. 

The real peaks that can be reduced without modifying the sound
are the short ones. Short in time means broad in the frequency
domain. So they don't correspond to spectral peaks. The way
to deal with them is using a fast lookahead limiter.

Multiband compression can be used to raise the average level.
What happens there is way too complex to be done manually by
looking at a spectrum. It works by raising the level of the
lower energy bands, not by assuming that the one with the
highest level is the one causing clipping. You still need a
peak limiter after the compressor. It also works because each
band can have its own attack/release setting (faster for higher
frequencies), which means that you can have more compression
without it becoming too obvious.

> What would you suggest as a good criterion to find or at least
> estimate such points?

'Points' meaning spectral peaks ? I'd suggest not trying to
achieve anything in that way at all. 

> Yes there are the ears and they are you starting point 
> always. But they can be fooled or not careful enough.

Believe me, I've seen _much_ more people being fooled by a
spectrum analyser than by their ears. 



More information about the Linux-audio-user mailing list