On 2/24/2015 12:10 PM, Brian Morrison wrote:
On Tue, 24 Feb 2015 16:32:21 -0000
FranK Turner-Smith G3VKI wrote:

Bandwidth reduction for one.
If you can encode and compress speech sufficiently then you can use
less bandwidth in transmission.

That's the bit I have trouble getting my head around. Back in the
1970s and 1980s digital transmissions used a much greater bandwidth
than their analogue equivalents. Sampling at 2.2 x max frequency x
number of bits plus housekeeping bits etc. etc.

Have a look at David Rowe's web site about Codec 2 and his work on it.

http://rowetel.com

Most of the codec development effort goes into voice modelling that
allows redundant information to be thrown away without making the
encoded speech sound too horrible when decoded. And on working out
which bits in the encoded frame need to be better protected and which
don't, this is especially important when considering what encodes
voiced and non-voiced speech and ensuring it doesn't get mixed up.

Other than uLaw/Alaw, voice for telephony is not compressed in the same
ways that a zip file is. As you say, they model the vocal tract and
send the parameters for the sounds that are to be produced along with
error information to make it more intelligible. For sounds that aren't
voice or voice like, they are reproduced poorly. This is why low bit
rate compression on cell phones doesn't convey music very well and
background noise messes up the intelligibility much more than with uLaw
or ALaw compression which are just ways of compressing the waveform
without knowing anything about the content.

--

Rick