Home Mitarbeiter Lehrportal für Studierende Forschungsprofil Publikationen Domestikations- und Säugetierforschung Neurobiologie & Hörforschung Stimmforschung Organisation von Forschung & Lehre Forschungswerkstatt 2001 - 02 - 03 - 04 Krankenversorgung Kooperationen & Links

Glottal to Noise Excitation Ratio (GNE) Einleitung The glottal to noise excitation ratio (GNE) is designed as an acoustic measure to assess noise in a pulse train that is typically generated by the oscillation of the vocal folds. It is based on the assumption that glottal pulses resulting from the collision of the vocal folds lead to a synchronous excitation of different frequency bands. Turbulent noise generated at a constriction, on the other hand, leads to an uncorrelated excitation. The synchronism is expressed by the correlations between envelopes calculated for the different frequency bands. The algorithm to calculate the GNE (more precisely to calculate gne3) looks as follows (for a more detailed description see [Michaelis et al. 1997]. linear-predictive inverse filtering of the speech signal to obtain glottal pulses (if present) bandpass filtering of the residual signal in the frequency domain by applying Hanning windows (3000Hz width) at different center frequencies calculation of the Hilbert envelopes for each frequency band in the frequency domain and back-transformation to the time domain calculation of correlation coefficients between the different Hilbert envelopes for lags in the range -0.3ms < t < 0.3ms (difference of center frequencies has to be at least 1500Hz) maximum correlation coefficient defines the GNE GNE is sensitive to broad band noise which was tested in a study using synthetic signals [Michaelis et al. 1997]. It reaches its maximum value of 1.0 if the envelopes in two different frequency bands are exactly the same. However, a general restriction for the generalization of results on synthetic signals lies in the appropriateness of the model used to generate the data. Therefore tests on real speech are equally important. The performance of the GNE has already been discussed for several case studies of various voice conditions [Fröhlich et al. 1997]. Generally, for real voices white or high-frequent noise might be present in the speech signal at a level at which it already affects the higher harmonics (due to the spectral tilt) while the lower harmonics still remain relatively unchanged. Since voice signals contain most energy at the low frequencies, in this case the signal periodicity - insofar as it existed - would be hardly affected. However, the correlation between the envelopes in the different bands would be diminished considerably. Therefore, a lower GNE would result. References [Fröhlich et al. 1997] Advances in Quantitative Laryngoscopy, 2nd "Round table", Erlangen (1997) [Michaelis et al. 1998] Glottal-to-Noise Excitation Ratio - a New Measure for Describing Pathological Voices acustica/acta acustica (1997)