Analyzing geospatial variation in articulation rate using crowdsourced speech data

Abstract

Most recent studies on the geographical distribution of acoustic features analyze comparatively few speakers and localities, both of which may be unrepresentative of the diversity found in larger or more spatially fragmented populations. In the present study we introduce a new paradigm that enables the crowdsourcing of acoustic features through smartphone devices. We used Dialäkt Äpp, a free iOS app that allows users to record themselves, to crowdsource audio data. Nearly 3,000 speakers from 452 localities in German-speaking Switzerland provided recordings; we measured articulation rates for these speakers using a metric based on duration intervals between consecutive vowel onsets. Results revealed distinct regional differences in articulation rate between major dialect regions and individual localities. The specification of 452 localities enabled analyses at an unprecedented spatial resolution. Results further revealed a robust effect of gender, with women articulating significantly more slowly than men. Both the geographical patterns and the effect of gender found in this study corroborate similar findings on Swiss German previously reported in a very limited set of localities, thus verifying the validity of the crowdsourcing framework. Given the application of this new framework, a large bulk of the discussion is devoted to discussing methodological caveats.