This appendix deals with optimization of the silence compression scheme, described in Annex B, in conjunction with the main body of ITU-T Rec. G.729, and Annexes A, C or H/G.729, when used for VoIP applications.
This appendix addresses in particular the following issues noted for Annex B:
1) Initialization of background noise statistics at the beginning of the call and updates of background noise statistics when the background noise characteristics change;
2) Early estimation of SID parameters which generates breathe-like noise in sharp-edge energy offsets;
3) Classifying of portions of very long and high-level tonal signals as "inactive speech";
4) Frequent changes between "inactive speech" and "active speech" for particular types of background noises;
5) Frequent SID update frames, which is undesirable for VoIP applications where the packet overhead information is considerably larger than the payload for such frames.
This appendix provides solutions for the issues noted for Annex B. This appendix is intended for VoIP systems, providing an improved quality over Annex B while maintaining bandwidth efficiency.
This appendix contains three electronic attachments. The first two consist of the modified source C code files for dtx.c and vad.c (of Annex B/G.729). The other is a set of test files that were used to test the algorithm in this appendix.