Comparing pitch distributions using Praat and R

Abstract

Pitch analysis tools are used widely in order to measure and to visualize the melodic aspects of speech. The resulting pitch contours can serve various research interests linked with speech prosody, such as intonational phonology, interaction in conversation, emotion analysis, language learning and singing. Due to physiological differences and individual habits, speakers tend to differ in their typical pitch ranges. As a consequence, pitch analysis results are not always easy to interpret and to compare among speakers. In this study, we use the Praat program (Boersma & Weenink 2015) for analyzing pitch in samples of conversational Finnish speech and we use the R statistical programming environment (R Core Team, 2014) for further analysis and visualization. We first describe the general shapes of the speaker-specific pitch distributions and see whether and how the distributions vary between individuals. A bootstrapping method is applied to discover the minimal amount of speech that is necessary in order to reliably determine the pitch mean, median and mode for an individual speaker. The scripts and code written for the Praat program and for the R statistical programming environment are made available under an open license for experimenting with other speech samples. The datasets produced with the Praat script will also be made available for further studies.