A Meta-Analysis: Acoustic Measurement of Roughness and Breathiness

Journal of Speech, Language & Hearing Research

Barsties v. Latoszek, B., Maryn, Y., et al. (2018).

Journal of Speech, Language & Hearing Research, 61(2), 298-323.

This meta-analysis investigates the correlations between auditory-perceptual and acoustic measures of roughness and breathiness in sustained vowels and continuous speech.

Not stated

1960 to 2013

Published and unpublished studies (not further specified)

33 (roughness); 34 (breathiness)

<p>Acoustic measures with higher correlation coefficients (weighted mean<em> r</em> &ge; .60) with auditory perceptual measures were considered to be more valid; that is, they better "quantify/objectify the subjective/perceptual concept" (p. 2) of roughness or breathiness.</p> <p>For roughness, this included the following as the most promising measures:</p> <ul> <li>Six spectral noise level (SNL) parameters: <ul> <li>100-5100 Hz (r = 0.91);&nbsp;</li> <li>100-8100 Hz (r = 0.90);&nbsp;</li> <li>100-2600 Hz (r = 0.89);&nbsp;</li> <li>100-3000 Hz (r = 0.85);&nbsp;</li> <li>2600-5100 Hz ( r= 0.82); and&nbsp;</li> <li>5100-8000 Hz (r = 0.75);</li> </ul> </li> <li>Second harmonic amplitude (H2A; r = 0.73);&nbsp;</li> <li>Pearson <em>r</em> at autocorrelation peak (RPK; r = 0.71);</li> <li>Normalized noise energy (NNE; r = 0.68*);</li> <li>Jitter Factor (r = 0.67);</li> <li>Glottal-to-noise excitation ratio (GNE) 1000 Hz (r = 0.65);</li> <li>Amplitude variability index (AVI; r = 0.63);</li> <li>Energy perturbation quotient (EPQ; r = 0.62*); and</li> <li>Smoothed pitch perturbation quotient (sPPQ; r = 0.61).</li> </ul> For breathiness, this included: <ul> <li>Natural log of period standard deviation (LNPSD; r = 0.76);</li> <li>GNE 3000 Hz (r = 0.71);&nbsp;</li> <li>Differences between the amplitude of Formant 0 and Formant 1 (F0-F1 [L0-L1]; r = 0.72);</li> <li>Relative energy level of high frequency noise (Hfno; r = 0.70);&nbsp;</li> <li>Cepstral peak prominence (CPP; r = 0.66*);</li> <li>Differences between the amplitude of the first and second harmonics (H1-H2; r = 0.66);</li> <li>Smoothed cepstral peak prominence (CPPs; r = 0.64*);&nbsp;</li> <li>Smoothed pitch perturbation quotient (sPPQ; r = 0.64);</li> <li>Harmonic-to-noise ratio from Dejonckere &amp; Lebacq (1987; HNR Dejonckere; r = 0.63)</li> <li>Amplitude perturbation quotient-5 (APQ5; r = 0.62);</li> <li>Normalized noise energy (NNE) 1000-5000 Hz (r = 0.61*); and</li> <li>Smoothed amplitude perturbation quotient (sAPQ; r = 0.60).&nbsp;</li> </ul> Asterisks represent heterogeneity in results.