Per Stromgren:
(Posted earlier in rec.audio.tech with no protests, I take that as an
agreement of sorts.)
The threads that are hovering around Mykes work with sound files are
very much a discussion of what the words mean. In order to put that
particular discussion to rest I invite the parties to come to an
agreement about the words. I propose the following descriptions of
these often discussed words:
*** Normalize - to scale an entire sound file in order for its loudest
sample to hit full scale (0db FS), or a tiny bit less. This is
performed by first looking up the loudest sample, and then multiply
all samples by the (constant) factor that takes the loudest up to a
fraction under 0dB FS. The sound file's average level is not taken
into account at all when calculating the gain, only its loudest
sample. The effect is like turing the wolume knob up.

You might want to differentiate between Peak Normalization and RMS
Normalization (as a Peak Normalization definition, yours is almost
perfect).

Peak Normalization: scaling the amplitudes of all frequencies in a
sound file such that the highest peak is of a chosen amplitude. Never
introduces clipping, except in the case of a previous hard limiting.
Increases both dynamic range and S/N ratio.

RMS Normalization: scaling the amplitudes of all frequencies in a
sound file such that the RMS (Root Mean Square) average (apparent
volume; usually takes into account an equal loudness contour during
initial RMS calculation) is of a chosen volume. Can clip unless
measures are taken to prevent this (limiting or compression).
Usually used to match apparent volume levels between two or more
sound files.

*** Clipping - what often happens to a signal that is too loud for a
system. An amplifier can't put out more than plus and minus rail
voltage; any signal louder than that is "clipped" to this voltage,
which sounds awful. The same thing happens when you hit 0dB FS in a
digital system; it can't put out more than "all bits on" (not always
to be taken literally). This maximum level is sometimes called clip
level.

I would add "for two or more consecutive samples" after "hit 0dB FS"
and a note that it sounds like total ass when audio clips. A note about
wrapping might also be in order.

*** Limiting - what you perhaps want to do to a signal in order for it
to fit under your systems clip level, but not being clipped. It can be
accomplished by adjusting the gain for the signal, when it is to reach
clip level. This can be done by hand or automatically. In any case you
will get a compression of the signal.

Hard limiting: forced 'clipping' at a level lower than 0dB FS.
Straight-line peak trim. Can sound very harsh.

Soft limiting: compression applied only to the highest peaks to prevent
clipping, usually before a normalization; "shallow" compression.

*** Compression - A non-linear transpose function where the gain is a
function of the signal level. Most often it is used to amplify soft
passages more than loud ones to avoid the soft ones to drown in noise,
but is also used as an effect in modern music. Compressing is also
used as part of other signal handling, Dolby NR for instance. The
effect of compression on music is that it sounds louder, because its
average level has gone up.

Looks good, but a note about decreasing dynamic range wouldn't hurt.
Might also want to add its opposite, Expanding, to the list.

Is there a process which simply ADDS a chosen value to each and every
positive sample and subtracts that value from each and every negative
sample (ie, not DC offset - I want to shift equally in both directions)?
And yes, I realize this would raise the noise floor considerably...

-StArSeEd

--
dchub://tsphub.dyndns.org:1979
IRC EFnet #smashing_pumpkins
Email: am
ICQ UIN: 1711589