[VoiceOps] T38 over Sip

[Alex Balashov]

Alex Balashov wrote:

> Like many other compression 
> schemes, it also shrinks the size of the data by referring via shorthand 
> to elements of a waveform table/model that approximate the quantised 
> value of a sample, but do not EQUAL it.  

Side note:

Actually, G.711 - and indeed, standard DS1 digital PCM bearer channels - 
do some "rounding" too.  It's called logarithmic companding.  The type 
of logarithmic companding is indicated by the "mu" (uLaw) or A suffix. 
The basic idea there is that the more median portion of the spectrum of 
human vocal capabilities is expressed with closer fit, but the more 
outlying portions more approximately, with deltas in the "steps" that 
grow as a logarithmic progression.

This is why hold music sounds very lopsided on a phone - even without 
the fancy variable bit-rate and adaptive codecs used by mobile 
equipment.  Some parts of it are very obviously clearer than others. 
The other - and perhaps even more significant - reason is the 3.1 KHz 
total bearer spectrum.  Good human hearing tops out at about 20 KHz. 
Most music relies a lot on the higher range - certainly, well above ~3 KHz.

However, T.30 terminals (fax machines) were designed to work with this 
logarithmic companding in mind.  Companding is not the same as "vicious 
compression."

It's roughly the same difference as between a raw PCM WAV rip of a song 
from an audio CD, and the MP3 format.  The latter gets you upwards of 
10:1 compression ratio without significant loss from the point of view 
of human subjectivity, but if you're trying to encode any data into the 
analog signal, forget it.

-- 
Alex Balashov - Principal
Evariste Systems
Web     : http://www.evaristesys.com/
Tel     : (+1) (678) 954-0670
Direct  : (+1) (678) 954-0671