CPPS Normative in Hindi Speakers Using a Mixed-Phoneme and an All-Voiced Text With and Without Voicing Detection Algorithm

Abstract

Acoustic analysis of voice has proven to be a cost – effective, easy and reliable method of instrumental voice assessment over the past few decades and Smoothened Cepstral Peak Prominence (CPPS) has been found to be the most sensitive acoustic measure for voice analysis across many research studies. Many authors in the past have recommend the usage of an all – voice sample recording for CPPS analysis that is predicted to yield lesser false positives. Alongside, two different types of algorithms are incorporated in various acoustic analysis softwares for CPPS analysis. The first type uses an Automatic Voicing Detection (AVD) algorithm which extracts only the voiced segments from the sample and computer CPPS values on the extracted voiced sample. Whereas, the second type computes the CPPS on the whole sample. Both algorithms have their own merits and demerits. Since CPPS is one of the most valid and reliable acoustic measures of voice quality, but there is no study in the literature that has found CPPS normative in Hindi speakers using continuous speech sample, a study to establish this was needed to be done. This study aims to establish CPPS normative for different sample stimuli, i.e., sustained vowel phonation samples (/a/, /i/, and /u/), a mixed-sentence stimulus (that contains both voiceless and voiced phonemes) and a voiced-sentence (that contains only voiced phonemes) in young adult Hindi speaking normophonic individuals with and without voicing detection algorithm. 92 participants within the age range of 19 – 30 years were recruited for this study. Samples of all participants were subjected to perceptual and acoustic screening to ascertain normal quality of voice. Final screened sample consisted of 37 females and 36 males. All the screened samples were subjected to CPPS analysis using Praat software. Mean CPPS values for /a/, /i/, /u/, mixed-sentence, and voiced-sentence were reported. Use of AVD yielded higher CPPS values in comparison to when AVD wasn’t used. Overall, male samples yielded higher CPPS values than female samples. Amongst sustained phonation samples, /a/ had highest CPPS values followed by /i/ and /u/ in both female and male subgroups. Similarly voiced-sentence yielded higher CPPS values in comparison to mixed-sentence in both female and male subgroups. Given the disadvantages of using Automatic Voicing Detection, and low intertext variability and better sensitivity of CPPS when computed upon an all – voiced text, computation of CPPS using an all – voiced text without the use of automatic voicing detection is suggested for clinical purposes. However, further sensitivity and specificity of the obtained CPPS normative should be calculated by computing CPPS on dysphonic individuals using the stimuli and methods used in this study.