Jan Rychter wrote:
>Does G.729 provide better voice quality than GSM?
>
>(a question for people who have tried both)
>
It depends. The bit rate of G.729 is a lot lower, so it starts with a
disadvantage. To overcome that, they made it a lot more complex and
tuned to the human voice. The result is for a single person talking
G.729 sounds pretty good. When there are other sounds mixed in, GSM
degrades more gently. The bit rate advantage of very low bit rate codecs
isn't much of an advantage in most VoIP work. A single audio RTP stream
carries a huge overhead, if the latency is kept low enough for a two way
conversation. 8kps vs 13.2kbps sounds like a big advantage. Add the RTP
overheads and the difference looks much smaller.
Actually, the GSM we use here hasn't been used in GSM networks for
years. The use either EFR (enhanced full rate) or half rate. EFR is some
sort of CELP based codec (I can't remember the details) running at the
same bit rate as the original GSM codec - same bit rate; higher quality
on a single voice; less tolerant of more complex sounds. Half rate is a
5.95kbps VSELP codec. I think the half rate codec can beat the 8kbps
G.729, but it depends a lot on the implementation. Nokia phones suck on
half rate. Some others sound pretty good.
Codec performance is difficult to compare, as circumstances affect the
results a lot. Just a little background noise can often make a big
difference. Codec developers spend a fortune on trials of new designs
before drawing any real conclusions about them. As a crude illustraion
of the problem, look at comments people post about iLBC. They range from
awful to excellent. Personally, I know I would have to use such a codec
a lot before making a meaningful comment.
Regards,
Steve