Real-time implementations and applications of the US Federal Standard CELP voice coding algorithm

The authors describes two real-time implementations of the US Federal Standard 1016) (USFS-1016 CELP (code excited linear prediction) V3.2 4800 b/s voice coding algorithm. A floating-point implementation for the TI TMS320C30/C31 and a fixed-point implementation for the TI TMS320C25/C51 are described. The floating-point system implements the full 1016 standard including a 512-vector stochastic codebook search and fractional pitch search. Results from subjective listening tests which show the implementation to perform as well or better than the USFS 1016 FORTRAN simulation are presented. A 7200-b/s land mobile radio (LMR) system that augments the floating-point vocoder with 2400-b/s channel error coding has also been implemented. The error detection and correction performance of the LMR system is presented for various channel scenarios. A miniature vocoder system is presented.<
>