Chapter 13
Tonal annotation
Stylization of complex melodic contours over arbitrary linguistic units
Julie Belião | Modyco, Paris Nanterre University & CNRS, France
This chapter presents SLAM, an algorithm for the automatic stylization and labelling of melodic contours, developed to process intonation in Rhapsodie. This algorithm has three basic specificities. First, the alphabet of melodic contours is directly derived from the speech signal. Second, complex melodic contours are described though a simple time-frequency representation. Third, melodic contours can be described over various linguistic segments that can be specified by the user; in other words, the tonal descriptions provided by the system can be used to present the intonation of segments of any size, larger than the syllable and of any type: prosodic, syntactic, and informational. Additionally, the system handles some specificities of spontaneous speech, such as speech turns and speech overlaps.
Article outline
- 1.Introduction
- 2.Intonation labeling
- 2.1Pre-processing: Speech segmentation and adaptive F0 smoothing
- 2.2Acoustic representation of melodic contours
- 2.3Symbolic representation of melodic contours
- 2.3.1Frequency representation
- 2.3.2Time representation
- 2.3.3Formal representation
- 3.Processing interactive speech
- 3.1Speech turns
- 3.2Speech overlaps
- 4.Preliminary experiment
- 5.Conclusion
-
Note