Using Praat for high-quality speech manipulation and illustration: recommended practices and demonstrations
Matthew B. Winn
University of Minnesota
Praat is widely used and freely available software that is designed with speech acoustics in mind. Despite its popularity, there is little standardization on the development of features and scripts, causing duplication of work across labs. This presentation has two goals: 1) highlighting a number of features that could be useful to kickstart such standardization, and 2) demonstration of scripts that could be useful for experimenters looking to create high-quality perceptual stimuli or classroom demonstrations. Recommendations will be given for variable assignment, documentation, increasing stimulus naturalness, and incorporating auditory scaling rather than linear interpolation for both frequency and intensity. Demonstrations will include manipulations of vowel formant contours, voice onset time, pitch contour, and fricative spectra, all using natural speech. Additionally, resources will be highlighted that allow users to easily compare outputs of sound modifications like spectral filtering. Examples will be given of data visualization within Praat to verify clean manipulations, as well as creating publication-quality images using R while handling objects from Praat.
PACS: 43.72.Ja speech synthesis and synthesis techniques
Acoustical Society of America
Acoustics Virtually Everywhere 2020
Abstract ID: 3474079
https://github.com/ListenLab/Praat
https://github.com/ListenLab/VOT