![]() PowerCepstrogram: To Table (cepstral peak prominences).In the next picture the trend line is of exponential decay type and consequently the peak prominence value has changed a little bit. Note that the first four lines in the script are only necessary to generate a PowerCepstrum of a part of a vowel.Ĭreate KlattGrid from vowel: "a", 0.3, 125, 800, 80, 1200, 80, 2300, 100, 2800, 0.1, 1000 Next picture of a PowerCepstrum with its straight blue trend line and its corresponding peak prominence value was generated with the following script. To be compatible with the past, a standard least squares line fit can also be chosen but it is much less precise than the other two because a least squares fit is much more influenced by the peak cepstral values than the other two. The "Robust" method corresponds to the incomplete theil regression and is computationally faster but somewhat less precise. The default method, "Robust slow", corresponds to Theil's robust line fit. Or, we could use an exponential model in which the background cepstral amplitudes decay in a non-linear fashion.ĭefines how the line that models the cepstrum background is calculated. The slope of this line will generally be negative because the background amplitudes get weaker for higher quefrencies. We can model it with a straight line as was done in Hillenbrand et al. Table 1 summarizes how the acoustic outcome measures mean voice SPL, mean f o, and smoothed cepstral peak prominence (CPPS) were analyzed. If you choose the "Least squares" fit method then it matters more.ĭefines how to model the cepstrum background. Praat was used to conduct acoustic analysis of the sustained vowel samples using a custom analysis script. In our analysis this value is not so critical if we use the robust fitting procedure. (1994) article was chosen as 0.001 s in order to reduce the effect of very low quefrency data on the straight line fit. The lower value for this range in the Hillenbrand et al. The quefrency range for which the amplitudes (in dB) will be modelled by a straight line. A pitchCeiling of 300 Hz will correspond to a lower quefrency of 1/300≈0.0033 seconds.ĭetermines how the amplitude and position of a peak are determined. The lower quefrency is determined as 1 / pitchCeiling and this value is in general more critical than the value of the upper quefrency which equals 1 / pitchFloor. The CPP measure represents how far the cepstral peak emerges from the cepstrum background.ĭetermine the limits of the quefrency range where a peak is searched for. The CPP measure is the difference in amplitude between the cepstral peak and the corresponding value on the trend line that is directly below the peak (i.e., the predicted magnitude for the quefrency at the cepstral peak). All rights reserved.Calculates the cepstral peak prominence measure (CPP) as defined by Hillenbrand et al. Clinicians may consider using CPPS to complement clinical voice evaluation and screening protocols.Īcoustic measures Analysis of Dysphonia in Speech and Voice Dysphonia Praat Smoothed cepstral peak prominence.Ĭopyright © 2017 The Voice Foundation. CPPS measures from both programs were significantly and highly correlated (r = 0.88, P < 0.001).Ī single acoustic measure of CPPS was highly predictive of voice disorder status using either program. CPPS measures derived from Praat were uniquely predictive of disorder status above and beyond CPPS measures from ADSV (χ 2(1) = 40.71, P < 0.001). Results showed acceptable overall accuracy rates (75% accuracy, ADSV 82% accuracy, Praat) and area under the ROC curves (area under the curve = 0.81, ADSV AUC = 0.91, Praat) for predicting voice disorder status, with slight differences in sensitivity and specificity. Relationships between CPPS measures from the programs were determined. Logistic regression and receiver operating characteristic (ROC) analyses were used to evaluate and compare the diagnostic accuracy of CPPS measures. Measures of CPPS were obtained from connected speech recordings of 100 subjects with voice disorders and 70 nondysphonic subjects without vocal complaints using commercially available ADSV and freely downloadable Praat software programs. This is a retrospective cross-sectional study. The purposes of this study were to (1) determine and compare the diagnostic accuracy of a single acoustic measure, smoothed cepstral peak prominence (CPPS), to predict voice disorder status from connected speech samples using two software systems: Analysis of Dysphonia in Speech and Voice (ADSV) and Praat and (2) to determine the relationship between measures of CPPS generated from these programs. ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |