60 Automatic Silence/Sonorant/Non-Sonorant Detection Based on Multi-Resolution Spectral Analysis and Anova Method

This paper presents an automatic method for Silence/Sonorant/Non-Sonorant detection based on multi-resolution spectral analysis (MRS) and ANOVA method. Our method is composed of two parts. The first is calculation of the MRS by combining several FFT of different lengths. The second is an automatic Silence/Sonorant/Non-Sonorant detection by calculating a classical statistics method such as Standard Deviation and Mean and a statistical test such as the Analysis of Variance ANOVA of each N frames of the MRS FFT. We used our Visual Assistance of Speech Processing (VASP) System applied to a corpus. This corpus was in French Belgian pronounced by Belgian speakers. The analysis of the obtained results shows that automatic Silence/Sonorant/Non-Sonorant detection based on MRS provides better results compared to classical spectral analysis of the corpora used. Classifications with ANOVA was better then the classical methods such as Standard Deviation and Mean.