Forums

Comfort Noise Generator

Started by kanhaiya October 8, 2004
Dear Experts

Well I've read a few previous posts on this aspect. The documents or
references I used were:

RTP Payload for Comfort Noise RFC3389.
Appendix II of G.711
Annex B of G.729

Well my query is regarding implementation of a codec independent CNG:

1) Which excitation is best for this CN (white or pink)?
2) I tried with only noise energy level transmission, and generating
pink noise at the decoder and scaling the level to the level
transmitted. Didn't seem to sound too good. Which implementation is
found to be decent/good for CN?
3) Has anyone tried or is it better with transmission of the
Reflection coefficients (spectral info) for better sounding CN?
4) If so what should be the order of the filter suggested? or less
costly in terms of MIPS?

Regards
kanhaiya
Could some one address my queries please? I'm sure there would be some
one who's worked on comfort noise generation

Regards
Kanhaiya

kichainrec@rediffmail.com (kanhaiya) wrote in message
news:<52a0c495.0410072301.39a76109@posting.google.com>...
> Dear Experts > > Well I've read a few previous posts on this aspect. The documents or > references I used were: > > RTP Payload for Comfort Noise RFC3389. > Appendix II of G.711 > Annex B of G.729 > > Well my query is regarding implementation of a codec independent CNG: > > 1) Which excitation is best for this CN (white or pink)? > 2) I tried with only noise energy level transmission, and generating > pink noise at the decoder and scaling the level to the level > transmitted. Didn't seem to sound too good. Which implementation is > found to be decent/good for CN? > 3) Has anyone tried or is it better with transmission of the > Reflection coefficients (spectral info) for better sounding CN? > 4) If so what should be the order of the filter suggested? or less > costly in terms of MIPS? > > Regards > kanhaiya
kanhaiya wrote:

>Could some one address my queries please? I'm sure there would be some >one who's worked on comfort noise generation > >Regards >Kanhaiya > >kichainrec@rediffmail.com (kanhaiya) wrote in message
news:<52a0c495.0410072301.39a76109@posting.google.com>...
> > >>Dear Experts >> >>Well I've read a few previous posts on this aspect. The documents or >>references I used were: >> >>RTP Payload for Comfort Noise RFC3389. >>Appendix II of G.711 >>Annex B of G.729 >> >>Well my query is regarding implementation of a codec independent CNG: >> >>1) Which excitation is best for this CN (white or pink)? >>2) I tried with only noise energy level transmission, and generating >>pink noise at the decoder and scaling the level to the level >>transmitted. Didn't seem to sound too good. Which implementation is >>found to be decent/good for CN? >>3) Has anyone tried or is it better with transmission of the >>Reflection coefficients (spectral info) for better sounding CN? >>4) If so what should be the order of the filter suggested? or less >>costly in terms of MIPS? >> >>Regards >>kanhaiya >> >>
If I remember correctly, the comfort noise feature they added to G.711 is actually the comfort noise feature originally defined in G.729. I also seem to recall that G.729 fully defines how comfort noise should be implemented. That is considered a fairly good comfort noise technique. Why not just implement that? One implementation should be in the G.729 reference code, although it unclear what rights you may/may not have to reuse that code in a real product. It isn't the lowest MIPs technique possible, but you need to decide what you want - cheap or good. I doubt you can have both. Personally, I think CNG sucks, even in the best implementations. It is, however, marginally better than using nothing at all. Regards, Steve
> If I remember correctly, the comfort noise feature they added to G.711 > is actually the comfort noise feature originally defined in G.729. I > also seem to recall that G.729 fully defines how comfort noise should be > implemented. That is considered a fairly good comfort noise technique. > Why not just implement that?
Yes I have seen the code, but they use the LSF's. I wanted one just using the Reflection coeffs as described int he Appendix II of G.711, which they say is cheaper in terms of MIPS.