Perception and acoustic measurement of good voice quality for radio

Abstract

PURPOSE: Significant differences between voices of radio performers and controls have been reported in previous literature. Despite this, listener responses to voices on radio are likely to vary as voices need to suit the station/product. This paper examined whether a) voices could be reliably categorised based on how good they were for radio and b) these perceptual categories could be predicted using acoustic measures. METHOD: Twenty-four male radio performers (mean age = 36 years, range = 20-52 years) and 24 agematched male controls performed the 'Rainbow Passage' as if presenting on radio. In Study 1, the voice samples were perceptually rated using a three-stage paired-comparison paradigm by 51 naive listeners. Reliability data (Cronbach’s Alpha/ICCs) were used in forming perceptual groups. In Study 2, the same voice samples were analysed for measures of: Fundamental Frequency, Long Term Average Spectrum, Cepstral Peak Prominence and pause/spoken-phrase duration. RESULTS: In Study 1, Good inter-judge reliability was found for the perceptual ratings of best 15 voices (good for radio group, 14/15=radio performers) however agreements for ratings of remaining 33 voices (unranked group) were poor. Discriminant function analyses in Study 2 showed that Standard Deviation of Sounded Portion Duration, Equivalent Sound Level and smoothed Cepstral Peak Prominence predicted membership of good for radio and unranked groups with moderate accuracy (R2=0.328). DISCUSSION: Although some voices on radio could be perceived and agreed upon as having good quality for radio, current acoustic analyses can detect only some of the acoustic signal properties of these voices. Further research on perception of good voice quality for radio and acoustic measures that sensitively reflect good voice quality is warranted.