Visual comparison of speaker groups

We describe a generic tool for visualising differences between two groups of speakers who produce a given word sequence. We do this by first time-aligning all recordings and then aggregating time-varying information within each group. By that, we can display prototypical loudness and tempo contours, and also spectrograms, together with information on variability and group effect size over time. An optional user-supplied segmentation (just needed for one of the recordings) can be used to relate local differences to individual phonemes. The system is validated with a group of speakers with Parkinson’s disease and an age-matched control group. It will be provided as an opensource software package to the community.