Statistische langzeitanalyse von sprechstimmen auf frequenzgruppenbasis Statistical longterm analysis of speech sounds based upon critical band width

Summary

The paper presents a new method of statistical longtime speech analysis based upon the filter band width of the ear, called critical bands. It's not the average power spectrum which we measure but rather the time during which a certain trigger level is reached within each frequency band. By the choice of different trigger levels we get some insight into energy distribution in every frequency group. The research comprises of normal male voices as well as female. The overall distribution shows differences between the two sexes due to correlations of filter band width and voice characteristics. The distribution shows 3 maxima in all cases, this means that frequency ranges with long durations for a certain trigger level alternate with those presenting only short durations. This structure is more obvious at higher trigger levels than at low ones. In the case of high trigger levels the first maximum is divided structure into 2 very fine maxima in the case of male, into 3 in the case of female voices. This is due to different fundamental frequences. Because of this correlation between filter band width and fundamental frequency on the one side and bandwidth and formant frequency on the other side, changes of the distribution in certain cases of speech and voice disorders can be predicted. Varied formant structure, as appears in the case of deaf persons, is characterised by small frequency ranges where certain trigger levels are reached. Noise components in oesophageal speech show long durations of even higher trigger levels in the upper frequency domain.

Zusammenfassung

In der vorliegenden Arbeit wird eine neue Methode der statistischen Langzeitanalyse vorgestellt, wobei die Filterwahl entsprechend der Frequenzanalyse des Ohres in Frequenzgruppen erfolgt. Dabei erhält man innerhalb jeder Frequenzgruppe die Gesamtenergie nach der Häufigkeit der einzelnen Pegel aufgeschliisselt. Die Gesamtverteilung weist bei gleichem Text charakteristische Unterschiede zwischen Männer- und Frauenstimmen, die sich aus der Wechselwirkung von Stimmanteilen und Filtern erklären lassen, auf. Auf Grund diesel Zusammenhanges lassen sich Änderungen in der Gesamtverteilung bei bestimmten Stimm- und Sprechstörungen voraussagen. Dies wird an Hand von 2 Beispielen vorgeführt.