[LAD] peak has nothing to do with loudness (check out EBU R128)

Jörn Nettingsmeier nettings at folkwang-hochschule.de
Mon Dec 20 18:27:28 UTC 2010

On 12/20/2010 03:49 PM, Paul Davis wrote:

>> The documentation of the natural drum sample library is quite good. The only
>> thing missing is the "loudness" of each sample in order to map each sample
>> to a velocity level from 0-127.
>> What would you  recommend in order to calculate the "peek" of each drum
>> sample automatically? Is there a library which could do this? I would also
>> be happy with a command line tool like this:
> sndfile-info /music/misc/onewayjam_organ.ogg
> Signal Max  : 0.805912 (-92.18 dB)    <<<<<<<<< HEADS UP!
> Works for any audio fle that libsndfile can read, which is just about anything.
> there is a complication: loudness is no identical to maximum sample
> value, but the relationship is good enough for government work, so to
> speak.

no, it's not. there is absolutely no meaningful relation whatsoever
between perceived loudness and peak value. that's why the EBU is (after
all those years) moving to k-weighted loudness as a metering standard
rather than QPPM. true, peak measurement used to be the norm, but it was
never good for anything (not even government work) except to avoid
clipping downstream.

florian camerer of orf presented a very enlightening talk at the
tonmeistertagung in leipzig this year, which i had meant to post earlier
but forgot. thanks for this reminder :)


it starts a bit boastful, but rightfully so, since it's in effect _the_
peace treaty to end the loudness war.

bottom line: those who have been using bob katz' k-system (for example
as implemented by fons adriaensen in jkmeter) don't have to change much...
but if you're interacting with radio stations regularly, you might want
to implement the EBU R128 standard, whose loudness measuring method is
slightly different.

More information about the Linux-audio-dev mailing list