Cepstral Peak Prominence in normophonic Irish-English speaking adults: The effect of gender, age, speech task segmental composition, recording conditions and CPP extraction method.

Abstract

This paper reports Cepstral Peak Prominence (CPP) values for normophonic English-speaking adults living in Ireland. A hundred vocally healthy adults (50 females, 50 males) aged 18-55 were audio recorded producing a series of speech tasks (sustained vowels, connected speech). Fifty-eight speakers in the 18-24 age group were recorded twice: remotely, using their mobile phones and onsite, in controlled recording setup. Forty-two speakers in the 25-55 age group were recorded only remotely, using their mobile phones. CPP values were extracted automatically, with and without voice activity detection (VAD), using a Praat plug-in. Linear mixed effect model analyses were conducted to establish the effect of speaker gender, age, speech task and recording conditions on CPP values. The results suggest strong significant effect of speech task. Sustained vowel [a] was found to have the highest CPP. In other speech tasks CPP values were found to decrease as the number of voiceless obstruents in them increases. We found moderate significant effect of gender: in sustained vowels only CPP_[Male] > CPP_[Female]. No effect of age was found. Recording condition had significant, but rather small effect, and its direction was different for CPP_noVAD (Onsite < Remote) and CPP_VAD (Onsite > Remote). CPP_VAD values are higher than CPP_noVAD ones regardless of the speaker gender. The choice of CPP extraction method affects mainly the speech tasks containing voiceless obstruents (CPP_VAD > CPP_noVAD). The findings reported here contribute to the growing body of knowledge about normative CPP values and may serve as a reference in clinical voice assessment.