Christopher R Watts1, Shaheen N Awan2, Youri Maryn3. 1. Texas Christian University, Fort Worth, Texas. Electronic address: c.watts@tcu.edu. 2. Bloomsburg University, Bloomsburg, Pennsylvania. 3. European Institute for ORL, Sint-Augustinus Hospital, Antwerp, Belgium.
Abstract
PURPOSE: This study aimed to investigate the relationship and reliability of cepstral peak prominence (CPP) measures from two acoustic software applications, Analysis of Dysphonia in Speech and Voice (ADSV) and Praat. METHODOLOGY: Flemish and English recordings of sustained vowels and connected speech samples were analyzed using ADSV and Praat. Correlational analyses and measures of the standard error of the estimate were applied to the vowel and connected speech data obtained from the two programs. RESULTS: Analyses revealed very strong relationships (eg, r > 0.88) between CPP measures derived from ADSV and those derived from Praat, regardless of context (vowel or connected speech) or language spoken. Average residual errors ranged from 0.55 to 1.1 dB for the prediction of Praat CPP data from actual observed ADSV CPP data, and average residual errors ranged from 0.57 to 1.58 dB for the prediction of ADSV CPP data from actual observed Praat CPP data. CONCLUSIONS: Measurements of CPP derived from ADSV and Praat manifested strong parallel-forms reliability. Although CPP data values obtained via these programs will be different owing to algorithmic processing differences, this study found that estimated CPP values derived using regression equations could be transformed between programs with relatively small predictive error, regardless of language. The strong measurement relationships indicate that CPP values from either program have a high degree of shared variance and may be expected to differentiate across a wide range of voice signal periodicity in a relatively similar fashion. This finding supports the use of either program in clinical use and voice science research.
PURPOSE: This study aimed to investigate the relationship and reliability of cepstral peak prominence (CPP) measures from two acoustic software applications, Analysis of Dysphonia in Speech and Voice (ADSV) and Praat. METHODOLOGY: Flemish and English recordings of sustained vowels and connected speech samples were analyzed using ADSV and Praat. Correlational analyses and measures of the standard error of the estimate were applied to the vowel and connected speech data obtained from the two programs. RESULTS: Analyses revealed very strong relationships (eg, r > 0.88) between CPP measures derived from ADSV and those derived from Praat, regardless of context (vowel or connected speech) or language spoken. Average residual errors ranged from 0.55 to 1.1 dB for the prediction of Praat CPP data from actual observed ADSV CPP data, and average residual errors ranged from 0.57 to 1.58 dB for the prediction of ADSV CPP data from actual observed Praat CPP data. CONCLUSIONS: Measurements of CPP derived from ADSV and Praat manifested strong parallel-forms reliability. Although CPP data values obtained via these programs will be different owing to algorithmic processing differences, this study found that estimated CPP values derived using regression equations could be transformed between programs with relatively small predictive error, regardless of language. The strong measurement relationships indicate that CPP values from either program have a high degree of shared variance and may be expected to differentiate across a wide range of voice signal periodicity in a relatively similar fashion. This finding supports the use of either program in clinical use and voice science research.
Authors: Marziye Eshghi; Kathryn P Connaghan; Sarah E Gutz; James D Berry; Yana Yunusova; Jordan R Green Journal: J Speech Lang Hear Res Date: 2021-10-29 Impact factor: 2.674
Authors: Meike Brockmann-Bauser; Jarrad H Van Stan; Marilia Carvalho Sampaio; Joerg E Bohlender; Robert E Hillman; Daryush D Mehta Journal: J Voice Date: 2019-12-17 Impact factor: 2.009
Authors: Matthew L Rohlfing; Daniel P Buckley; Jacquelyn Piraquive; Cara E Stepp; Lauren F Tracy Journal: Laryngoscope Date: 2020-09-19 Impact factor: 2.970