Min Shu1, Jack J Jiang2, Malachi Willey3. 1. Department of Otolaryngology Head & Neck Surgery, Eye, Ear, Nose and Throat Hospital, Fudan University, Shanghai, China. 2. Department of Otolaryngology Head & Neck Surgery, Eye, Ear, Nose and Throat Hospital, Fudan University, Shanghai, China; Division of Otolaryngology-Head and Neck Surgery, Department of Surgery, University of Wisconsin School of Medicine and Public Health, Madison, Wisconsin. Electronic address: jjjiang@wisc.edu. 3. Division of Otolaryngology-Head and Neck Surgery, Department of Surgery, University of Wisconsin School of Medicine and Public Health, Madison, Wisconsin.
Abstract
OBJECTIVE: To investigate the effects of the moving window method on acoustic measures and discrimination ability between normal and disordered voices. METHODS: Fifty-three normal voices and 50 disordered voices were recruited. Three selection methods, the moving window method, the mid-vowel method, and the whole vowel method, were applied to each raw audio signal to determine the most stable segment of each signal. Acoustic parameters such as percent jitter, percent shimmer, signal-to-noise ratio (SNR), cepstral peak prominence (CPP), and correlation dimension (D2) were calculated. The Wilcoxon test was used to compare the stability of these segments across different methods. An artificial neural network was used for estimating how well disordered voices were discriminated from normal ones. RESULTS: Segments selected using the moving window method were more stable than those selected using the other two methods, meaning lower perturbation and nonlinear dynamic measurements as well as higher SNR and CPP values. The discrimination accuracy rate for the moving window method was 91.90 ± 8.73%, whereas the mid-vowel method and the whole vowel method were 72.34 ± 12.94% and 70.34 ± 5.24%, respectively. CONCLUSION: The moving window method is capable of providing a more stable audio segment and can discriminate disordered voices from normal ones more effectively.
OBJECTIVE: To investigate the effects of the moving window method on acoustic measures and discrimination ability between normal and disordered voices. METHODS: Fifty-three normal voices and 50 disordered voices were recruited. Three selection methods, the moving window method, the mid-vowel method, and the whole vowel method, were applied to each raw audio signal to determine the most stable segment of each signal. Acoustic parameters such as percent jitter, percent shimmer, signal-to-noise ratio (SNR), cepstral peak prominence (CPP), and correlation dimension (D2) were calculated. The Wilcoxon test was used to compare the stability of these segments across different methods. An artificial neural network was used for estimating how well disordered voices were discriminated from normal ones. RESULTS: Segments selected using the moving window method were more stable than those selected using the other two methods, meaning lower perturbation and nonlinear dynamic measurements as well as higher SNR and CPP values. The discrimination accuracy rate for the moving window method was 91.90 ± 8.73%, whereas the mid-vowel method and the whole vowel method were 72.34 ± 12.94% and 70.34 ± 5.24%, respectively. CONCLUSION: The moving window method is capable of providing a more stable audio segment and can discriminate disordered voices from normal ones more effectively.