Not all of this is entirely true.
> > I am still looking for a half decent program that will allow
> > you to people to talk for free over the internet and works with
> > modems.
> I worked for a time at a compnay which produced Voice over IP routers. It
> is hard enough to get decent voice quality over a single 64kbit/s link let
> alone on the internet.
64kb/s is enough bandwidth for one telephone quality voice transmission,
even just PCM coded, possibly at most 16 bits per sample. You could get
away with 8 bits even.
> There are numereous problems:
> 1. The delay in transmission of packets. This is high for internet
> 2. Packets sent out of order.
> 3. Dropped packets are not resent.
This is all true, however, the way speech is parameterized and processed
in resynthesised in internet phone systems means that lost packets now
and then will not degrade the transmission much as far as your ear is
We use UDP specifically because of it's speed. Of course, with UDP you
don't get resent packets, but in speech coding what good is resending a
packet that's already lost and therefore increasing the delay in output
if it doesn't make much difference perceptually anyway?
Quote:> To over come these problems there needs to be a large playback buffer at
> the recieving. I'd image that the buffer would be so large as to
> introduce a excessive playback delay >1 second. Compare with satellite
> delays of 0.5 seconds.
A buffer does help, but because of the delay aspect, it is not the main
draw card of an internet phone. The way the speech is repesented in
*information terms*, is the key to success. There is a lot of redundancy
in a speech signal.
> 4. Speech packets must be prioritised. Many routers do not prioritize.
That's not so important anyway for the same reasons above.
> 5. the UDP/IP packet headers must be compressed reduce the bandwidth by
> half. This only functions over a Point-to-point link where both sides
> understand the compression protocol. cf. Uncompressed voice requires about
> 16 kbit/s; compressed about 8 kbit/s. (voice data is a 24 byte packet
> every 30ms add a 30 (?) byte UDP/IP header and you double the bandwidth
You need to understand that a speech coder (the guts of an internet
phone) will produce data for transmission which has the reduncancies of
the speech signal removed (or at least as much as possible). Therefore
you cannot expect to compress the speech data one iota. This is all
based in information theory. The only overhead worth worrying about
after speech paramterization is the header.
Typical coding rates for speech are from about 1200 bits/s upward. And
that is for tried and tested techniques. Research is heading toward 800
bits/s now. But the point is 1200-2400 bit/s is a typical, moderate
Quote:> Basically it is not possible to have decent speech quality over the
Of course it is possible. I personally have used an internet phone over
a 56kb/s modem and the internet. People use these things all the time
now -- with reasonable speech quality.
> David Vrabel
> Engineering Undergraduate at University of Cambridge, UK.
Matthew B. Kennedy
Research Centre in Speech, Audio and Video Technology
Queensland University of Technology