Thursday, June 02, 2011

Adverse Effect of Noise on Voice Perturbation Estimates: A Comparison of Three Voice Analysis Programs

doi: 10.5336/medsci.2010-19461
Turkiye Klinikleri J Med Sci 2011;31(2):427-31
Mehmet Akif KILIÇ MD, Haldun OĞUZ MD, Fatih ÖĞÜT MD.
Objective: The aim of this study is to investigate the influence of noise on voice perturbation outputs obtained by three different voice analysis programs, and to show the software more immune to noise.
Material and Methods: For this purpose, 10 natural and 10 semi-synthetic voice samples were recorded. They were mixed with environmental noise and white noise. The unmixed and mixed signals were analyzed by Dr. Speech, MDVP and Praat. Three frequency perturbation outputs (jitter percent, relative average perturbation and pitch perturbation quotient) and two amplitude perturbation outputs (shimmer percent and amplitude perturbation quotient) were obtained. The unmixed and mixed signals were compared. Correlations were calculated between natural unmixed and natural mixed ones.
Results: The frequency perturbation outputs obtained by Dr. Speech and Praat seemed to be less affected by noise. However, the amplitude perturbation values measured by three systems were severely affected by noise, but Praat's performance found mildly better than the others'. Correlation analyses for 10 items (five parameters x two noise situations) revealed that there were perfect correlations (r= 1.000) regarding six items for Praat, three items for Dr. Speech and one item for MDVP.
Conclusion: As a conclusion, it was decided that Praat was the most immune one to noise among three analysis systems, followed by Dr. Speech and MDVP, respectively.