This repository contains data and code as well as instructions to reproduce our experiments in [1].
###################
###################
In order to reproduce the experiments described in [1] you need to firstly download and install the following software:
- Weka Data Mining software [2] from the following link http://www.cs.waikato.ac.nz/ml/weka/downloading.html
- Praat for feature annotation [3] from our repository https://github.com/monikaUPF/featureAnnotationforPraat
NB: The extension of Praat for feature annotation currently works under Linux
###################
###################
The following data are made available from our corpus on read speech:
- Binary files: intensity and pitch objects created using Praat from raw speech files are included in binary_intensity.zip and binary_pitch1/2/3.zip.
- TextGrids: segmentation into words and thematicity spans are includid in thematicity.zip.
- Arff files: the sentence span database (SSD) and the thematicity span database (TSD) are included in arff.zip.
###################
###################
The whole pipeline consists in 7 modules. Except for module 1 that is written in python, the rest are Praat scripts. Scripts for extraction of acoustic parameters, feature annotation and exporting to txt format consist in 6 modules (from 2 to 7). These modules require the extension of Praat for feature annotation (see [3]) to run. They do not run on standard Praat. These modules include the improved intensity detection module presented in [4]. The modules need to be executed in order (from 1 to 7).
###################
###################
In order to reproduce our experiments from the beginning you need to follow these steps:
1. download and install the required software;
2. download speech data and praat scripts;
3. run the 5 modules on the data;
4. create the arff files or use the ones we provide;
5. perform the classification experiments on Weka explained in the publication [1]
#####################
#####################
No raw speech files are made available in order to comply with EU and National data protection legislations. Voice data are considered sensitive data as they may be used to identify individuals. The binary files made available in this repository comply with anonymization ans pseudonymization requirements from this legislation as it is not possible to reconstruct the original voice only from these binary files. On the other hand, binary files are enough to reproduce our experiments from the scientific point of view.
#####################
#####################
If you use the material included in this repository or modify the code please cite the following publication:
- [1] Domínguez, M., Burga, A., M. Farrús, and L. Wanner (2018). Compilation of Corpora to Study the Information Structure-Prosody Interface. Proceedings of the Language Resources and Evaluation Conferece. Miyazaki, Japan. pp 4030-5.
Further references:
- [2] Eibe Frank, Mark A. Hall, and Ian H. Witten (2016). The WEKA Workbench. Online Appendix for "Data Mining: Practical Machine Learning Tools and Techniques", Morgan Kaufmann, Fourth Edition, 2016.
- [3] Domínguez, M., I. Latorre, M. Farrús, J. Codina and L. Wanner (2016d). Praat on the Web: An Upgrade of Praat for Semi-Automatic Speech Annotation. In Proceedings of the 25th International Conference on Computational Linguistics, Osaka, Japan.
- [4] Domínguez, M., M. Farrús and L. Wanner (2016c). An Automatic Prosody Tagger for Spontaneous Speech. In Proceedings of the 25th International Conference on Computational Linguistics, Osaka, Japan.