The present state of digital vocoding technique: A diagnostic evaluation

Eight state-of-the-art pitch-excited digital vocoders (2400 bits/s) were evaluated with the Diagnostic Rhyme Test, a two-choice test of the discriminability of six elementary attributes of consonant phonemes. All of the systems tested performed adequately with respect to the attributes, voicing, nasality, sibilation, and compactness. None, however, permitted adequate discrimination of the attributes, sustention (friction versus plosion) and graveness. All systems reduced the gross discriminability of these attributes and significantly biased listeners perception of the states of one or both. It is hypothesized that these deficiencies stem primarily from inadequate spectrum sampling rate (i.e., frame rate) rather than inadequate "spectrum resolution."