[FFmpeg-soc] [soc] G.723.1 Encoder: Compute LPC coefficients

[Justin Ruggles]

Mohamed Naufal wrote:

> +/**
> + * Estimate autocorrelation of the input vector.
> [...]
> + * Use Levinson-Durbin recursion to compute LPC coefficients from
> + * autocorrelation values.
> [...]
> + * Hamming window coefficients scaled by 2^15

Are these functions required to be bit-exact as defined by the
specification or are they implementation-dependent?  I ask because they
are quite similar to our existing autocorrelation and levinson-durbin
functions (though obviously 16-bit fixed-point rather than
floating-point).  Also, is a Hamming window required or can some other
window function be used?

-Justin