Prosodic Annotation of oral archives

Abstract

Prosodic analysis of oral archives is often hampered by adverse recording conditions, making popular pitch tracking algorithm fail in many cases, so that gathered data may be useless. To address these limitations and allow prosodic research to be conducted on almost any kind of ancient or recent recordings showing less than optimal acoustic conditions, several dedicated functions have been integrated in the software program WinPitch. A first set of these functions allow the user to apply alternate pitch tracking algorithms among 7 available on selected signal segments, as some algorithms may perform better than others. If this process appears still unsatisfactory, a set of easy to use graphical commands are available to directly annotate pitch curves graphically, relying on an underlying narrow-band spectrographic display, whose frequency scale is automatically matched with the fundamental frequency curve scale. Piecewise linear graphic lines can be adjusted to melodic movements of any complexity as displayed on the spectrogram. Furthermore, the phonetic or phonological categories of these user placed annotations can be automatically labelled according to predefined classes, such as those available in ToBI notation system, or as melodic contours indicating dependency relations between stress groups.