7th International Conference on Spoken Language Processing
September 16-20, 2002
This study examines the speaker’s influence on mean phone durations. As long as speech rate variation is present, the result of such a study would be trivial because every speaker has a particular speech rate that naturally modifies phone durations. Therefore, in order to eliminate its influence on phone duration, we developed a normalization procedure which evens out the local variability of speech rate, and then applied it to a large database of spoken German. As would be expected, general linear model statistical analysis (GLM) showed that speech rate normalization strongly reduced the variance explained by the factor ‘speaker’. Nevertheless, the variance explained by the interaction between ‘speaker’ and ‘phone type’ remained constant. Consequently, each speaker has individual intrinsic phone durations.
Bibliographic reference. Pfitzinger, Hartmut R. (2002): "Intrinsic phone durations are speaker-specific", In ICSLP-2002, 1113-1116.