Gabriel, you wrote,
> Dan, what is "acceptable" changes over time, depending on what can be rea=
sonably expected in terms of methodology. Now it is so easy and cheap to re=
cord in uncompressed formats that for most situations there is no good reas=
on to accept the possibility of compression artifacts.
Agreed, I just wanted to point out that thinking "compression =3D crap" or =
"compression =3D not usable for science" is unnecessarily extreme.
> Of course, that doesn't exclude the use of compressed audio in science co=
mpletely, but as general rule I would advise people to avoid it, unless you=
really know your way around compression algorithms and can demonstrate tha=
t your measurements are not affected.
Yep.
> But even then: recordings are often also used for purposes other than jus=
t spectral analysis. Playback experiments, for example, where previously re=
corded sounds are presented to animals to test for a response. What if Laur=
a or someone else later wants to use her recordings for this? Mp3 and atrac=
are based on characteristics of human auditory perception, but who knows w=
hat a frog or bird hears? We cannot assume it is the same as a human (in fa=
ct we know it isn't), so lossy compression algorithms should really be avoi=
ded.
Agreed. Compression should only be used for extra-long monitoring or for fi=
nal distribution by download.
For the newbies I want to explain that "compression" means two different th=
ings in audio today. Depends on the context of the word.
1) Limiting and Compression: methods for reducing dynamic range by raising =
lower-level signals and turning down higher-level signals. Should be avoide=
d in field recording. Limiting and Compression are production tools used wh=
en elements are being mixed, i.e. making a film soundtrack or podcast.
2) Data compression: (MP3, AAC, etc.) also called "lossy compression" to em=
phasize the fact that the signal is altered in order to reduce its file siz=
e, without changing the volume or dynamics. Usually used for downloading. I=
t can be transparent at higher data rates (256K, around 1/5 compression) bu=
t gets ugly at lower rates like 128K.
Why do cell phones often sound awful? Low-rate data compression.
-Dan
|