Vocoder Performance on Voiceless Fricatives
C. Karis, Calvin R. Howard, Ladislav O. Dolanský · The Journal of the Acoustical Society of America · 1963
A series of intelligibility experiments has been conducted, utilizing an 18-channel vocoder-multiplex system. The test material consisted of CVC nonsense syllables containing the fricatives /s/, /∫/, /ɾ/, and /θ/. Listener-test results demonstrated that vocoder performance on these speech samples was poor. The low intelligibility was found to be primarily attributable to erroneous listener responses to prevocalic fricatives. Although classical ambiguities such as /f/-/θ/ confusion were present, there was, in addition, a predominance of voiced-fricative responses to the initial fricative. The possibility that this was due to improper voiced-unvoiced vocoder decisions was ruled out by a test in which only unvoiced excitation was used in the synthesizer. Voiced responses were still in evidence, indicating that these errors were due primarily to spectral-energy distributions. Consequently, the properties of fricatives and fricative-vowel transitions were investigated in detail. The results of (a) sonagram analysis of the samples, (b) digitized power-spectrum information provided by the AFCRL digital speech processor, and (c) waveform analysis are discussed. [The research reported in this paper was supported by the U. S. Air Force Cambridge Research Laboratories, Office of Aerospace Research, under Contract AF 19(628)-389.]