Test verification of LD-CELP: an objective measurement approach for a non-bit exact standard

A verification test procedure for the low-delay code-excited linear predictive (LD-CELP) speech coder which is expected to be adopted as CCITT Recommendation G.728 for universal applications of speech coding at 16 kb/s is presented. Unlike previous CCITT G-Series Recommendations, LD-CELP is not a bit exact specification. It is written for implementation in 32 bit floating point arithmetic on any of a number of commercially available floating point digital signal processor chips or for simulation in a high-level language such as C or Fortran. In spite of the differing precision of these implementations and simulations, it has been possible to achieve interoperability with a high degree of fidelity among all of them. The motivation for creating an objective measure-based verification procedure, the methodology for creating test vectors, and some results for the test vector set created when applied to five different implementations or simulations are discussed.<>