Short-term spectral characteristics of the EGG signal in different registers
Andreas Selamtzis, KTH
This study is concerned with the recognition of different registers or phonatory states based on electroglottographic measurements. Ten amateur choir singers performed two tasks: a) Production of upward and downward glissandi including a phonatory state shift from modal to falsetto (or inverse) b) Deliberate phonatory state shift while keeping F0 as constant as possible. The EGG data was processed by means of phase portrait reconstruction and parameterization of the phase portrait, cycle-by-cycle, by a complex DFT. For each vocal production a running Sample Entropy estimate was computed to track the moment of the phonatory state shift. The data corpus was analyzed to find correlations between the spectral components of the phase portraits and the phonatory state. Correlations as high as r=0.7 were found for the harmonic amplitude ratios H2/H1 and H3/H1, suggesting that they can be used successfully for phonatory state classification and tracking, e.g., in a real-time voice range profiling (VRP) system.