Cutting down on manual pitch contour annotation using data modelling

Yuki Asano, Michele Gubian, Dominik Sacha


When experimental studies on intonation are based on large data sets, manual annotation of F0 contours using predefined categories such as a ToBI (Tones and Break Indices) system is tedious, costly and difficult to provide reliability. We present two data-driven modelling techniques that provide visual and quantitative maps of the F0 contour data set. The maps can be used to determine which ToBI categories are present in the data and in what proportions. Importantly, parts of the map that are homogeneous enough, i.e. they contain only one ToBI category, can be directly labelled without involving manual annotation, hence cutting down the overall costs of annotation. The modelling techniques will be evaluated on a small data set where a complete manual ToBI annotation was carried out, hence providing a ground truth for the evaluation.


DOI: 10.21437/SpeechProsody.2016-58

Cite as

Asano, Y., Gubian, M., Sacha, D. (2016) Cutting down on manual pitch contour annotation using data modelling. Proc. Speech Prosody 2016, 282-286.

Bibtex
@inproceedings{Asano+2016,
author={Yuki Asano and Michele Gubian and Dominik Sacha},
title={Cutting down on manual pitch contour annotation using data modelling},
year=2016,
booktitle={Speech Prosody 2016},
doi={10.21437/SpeechProsody.2016-58},
url={http://dx.doi.org/10.21437/SpeechProsody.2016-58},
pages={282--286}
}