In QCELP-8 (IS96A) speech coder, there is a process to determine rate
for current frame of speech. Determination of the rate is based on 3
threshold of energy and average energy of current frame (R0).
Formulation to count threshold energy is a function of background
noise of current frame i, Bi. But I am confuse with this meaning,
because from a simulation (source code in C), Bi for initial frame or
frame 1 (B1) is set to a value (the value is HIGH_THRESHOLD_LIM =
5059644, where is this value come from ?). If the meaning of
background noise is energy of noise that lies in the speech signal,
why it is set to a value. Why don't we calculate this noise energy
directly ? Please give me an explanation about this ...
Actually, I am still learning in this speech coding, so I meet some
problems here. The problem may be simple for you, I don't understand
why we flatten the spectrum of speech signal, with the use of
preemphasis filter ?
Thanks for your answer.