FCC doing it again...

[Kemp at DOCKMASTER.NCSC.MIL]

Bill Poser writes:
 > Scot E Wilcoxon writes:
 >> A 3KHz voice transmission requires about 6000 bits per
 >> second (OK, maybe 4800bps?)
 >
 > 6000 bps for a 3Khz [sic] bandwidth means only 1 bit resolution.
 > 1 bit speech is surprisingly intelligible, but the quality is very
 > poor.  The telephone system uses 8 bits per sample, as do the
 > cheapo ADDACS you can buy for personal computers.  This gives a
 > theoretical dynamic range of 48 dB.  (For speech research we try
 > to get at least 12 bits resolution - good research quality ADDACs
 > give a theoretical 16 bits.)  So the data rate for telephone speech
 > is 48K bits per second.

Scot is right.  Bill's facts are correct but not his conclusions.
Speech with a bandwidth of 3.6 KHz can be transmitted with very good
quality at 4800 bps, using an algorithm known as Code Excited Linear
Prediction.  In fact, the Federal Government is in the process of
ratifying Federal Standard 1016 specifying the details of this method of
digital speech coding at 4800 bps.

The telephone companies presently use ADPCM coding at 32 Kbps for most
of their trunks, but work is underway on a low-delay CELP at 16 Kbps.
Speech coding at 4.8 - 8.0 Kbps will be used first on digital cellular
circuits where bandwidth is extremely tight.

Bill Poser's comments on bit rate and distortion apply only to straight
PCM, not speech compression systems.


Dave Kemp <Kemp at dockmaster.ncsc.mil> "My sister is a yahoo"