During the last years new promising techniques have emerged for voice coding at bit rates around 5 Kbps and below. These techniques include the Code Excited Linear Predictive vocoder (CELP) and the Multi Band Excited vocoder (MBE). The objective of this experiment was to select and specify a voice coding algorithm that could be used within the "Voice Communication Assembly" of the Hermes space plane. Because the basic principles of both vocoders are very different (analysis-by-synthesis and parametric), only a subjective intelligibility test offers a fair comparison. For this experiment, SAIT has implemented a CELP-based vocoder and a MBE-based vocoder using common design constraints (real-time full-duplex operation on a DSP TMS320C30 processor, 5 Kbps data stream including protection, and robust against up to 10"^ burst BER). We performed extensive intelligibility tests on these systems. Our objective was to quantify the performance of the systems with clean and noise corrupted speech. These intelligibility tests were based on nonsense words (CVC or Consonant-Vowel- Consonant) and sentences (MOS and Speech Reception Threshold tests). The diversity of the tests makes it possible to validate them. It also simplifies the interpretation of the results.
Bibliographic reference. Wery, B. W. M. / Steeneken, Herman J. M. (1993): "Intelligibility evaluation of 4-5 kbps CELP and MBE vocoders: the hermes program experiment", In EUROSPEECH'93, 67-70.