Data labeling and sampling effects in harmonics-to-noise ratios

The harmonics-to-noise ratio (HNR) was proposed for measurement of noise in sustained vowels. A generalized algorithm is described here for removing the dependence on the data offset and for reducing the influence of jitter and shimmer. In addition, a new algorithm is proposed that accommodates periodic perturbations. Formulations are presented that can be computed in a single pass through the data under certain assumptions, thus simplifying the implementation of an efficient analysis program. A mathematical analysis of the effect of small errors in pitch-period demarcation is presented. The analysis indicates that the sensitivity to demarcation errors depends on the preponderance of high-frequency components, and significant underestimation is predicted for /a/ and /i/ data. A method for correcting quantization and demarcation errors in pitch-period markers is described. The method is shown to be effective at reducing demarcation errors for all but the most severely perturbed waveforms. Examples of the use of this method for HNR estimation are provided. Finally, a comparison with recent work by Milenkovic J. Speech Hear. Res. 30, 529-538 (1987)] is presented.