A comparison of US Government standard voice coders

The new US Government standard 4800-b/s code-excited linear predictive (CELP) coder is compared with the 2400-b/s linear prediction (LPC-10e) and 16000-b/s continuously variable slope delta modulation (CVSD) Federal Standard voice coders. This comparison includes speech quality and intelligibility in quiet and noisy environments/channels, speaker recognition, analog and digital implementation requirements, tandem coding, and coding delay. CELP is shown to offer dramatic improvements in speech quality relative to LPC-10e and CVSD. Although CELP requires significantly more computation than CVSD or LPC-10e, a version can be implemented on a single, new-generation, digital-signal-processing (DSP) chip. CELP provides interoperability among various computationally complex implementations, and CELP's robust performance tolerates a wide range of speaker variation, background noise conditions, analog impairments (i.e. nonflat microphones and channels), tandems, and errors. These features make CELP ideal for many speech applications.<>