VAD on AMR frames

Started by July 20, 2009
hi all,

I am implementing Voip App on nokia devices.
I have API to AMR Decoding.
I want to implement my own VAD. the reason is that it seems that the AMR VAD doesn't give good performence.
In my app, I need to close down delays from the channel and I want to know which frame to throw away.

also, Its always nice to save some BW by sending less frames.

so, what I have at the TX side is both the encoded AMR frame and I decode the AMR and get PCM.

on the PCM I can calculate the energy/zero crossing.
I guess I need to set threashold to decide whether its voice frame or silence. how can I set this threashold?

I think of using MAX filter (length ~ 4 frames? 80ms) on the energy to avoid the changing of the energy.

any idea on this subject?