hi,
was looking at how speex handles lost frames, in
particular the function nb_decode_lost(). (v1.1.6)
read in the speex archives the lost frames were
"interpolated". it seems to me that the parameters
were copied from previous frame and an exponential
attentuation is applied to the excitation (with more
stochastic component) as below
exc[i]=pitch_gain*exc[i-st->last_pitch] +
fact*sqrt(1-pitch_gain)*speex_rand(innov_gain);
and eventually, the frame will be silent as mentioned.
don't suppose any future frame is involved?
rgds,
tk