I just wanted to ask if there is any known problems/issues with the quality of the audio when you normalize it close to one. To be more exact I normalize some buffers with the command "normalize 0.999999". I've been told by someone that it would be better if I would normalize close to 0.85. So is there any known quality issues when you normalize close to 1.0?

normalization

Depending on exactly what you're doing with the audio beforehand you might want to make sure you will not get inter-sample peaks, which can cause problems further on down the line.

Hey, thanks for the answer. I don't think I am really familiar with the "inter-sample field". I do know what generally goes on but I am not sure how to avoid introducing quality problems (caused by inter-sample peaks). If you have any suggestions and/or references to suggest please let me know.

as long as the volume change happens in floating point, which

is always the case in msp, there is no loss in quality which would

but generally you might run into quality problems when using

this is 1. a problem when mixing many channels and 2. when

the mixing issue is very present in maxmsp, because in opposite

to protools or cubase summing happens also "only" in 32 bit float.

where on a soundcraft mixer you can get quite a good sound from

going into red when summing, on a digital workstation you are

well advised to not put 50 channels at an average of +3db (with

peaks much higher) and then write a file to disk which has a peak of

the headroom of a summing engine in such software is limited and

if you hit the limit, single samples will be wrong, which sooner or later

normalizing to 1.0 is something you can only theoretically do, in

practice there is a big issue with cheap DA converters when the

good music which good mixed sounds good at -6db too, and with

-2 bits even the cheapest 16 bit converter will not produce

as long as the volume change happens in floating point, which
is always the case in msp, there is no loss in quality which would
be worth to worry about.

but generally you might run into quality problems when using
too loud samples:

this is 1. a problem when mixing many channels and 2. when
applying dithering some final song.

the mixing issue is very present in maxmsp, because in opposite
to protools or cubase summing happens also "only" in 32 bit float.

where on a soundcraft mixer you can get quite a good sound from
going into red when summing, on a digital workstation you are
well advised to not put 50 channels at an average of +3db (with
peaks much higher) and then write a file to disk which has a peak of
what you called 0.999 above.
the headroom of a summing engine in such software is limited and
if you hit the limit, single samples will be wrong, which sooner or later
make a recording sound "cold".

normalizing to 1.0 is something you can only theoretically do, in
practice there is a big issue with cheap DA converters when the
signal is that loud.

good music which good mixed sounds good at -6db too, and with
-2 bits even the cheapest 16 bit converter will not produce
any errors.

Your comment was really enlightening. I myself checked some things about inter-sample peaks and you are just right. Thanks again.

Normalization