SpeechCollectR: An R package for web-based speech data collection

Abstract

To accommodate web-based speech production and perception data collection during the COVID-19 pandemic, we developed the R package, SpeechCollectR (https://github.com/abbey-thomas/speechcollectr), designed for speech scientists familiar with R [R Core Team, R: A Language and Environment for Statistical Computing (R Foundation for Statistical Computing, Vienna, 2021]. SpeechCollectR employs existing R functions for acoustic analysis, allowing researchers to collect, visualize, and analyze speech production and perception data with a single program. SpeechCollectR lets researchers construct interfaces that embed experimental tasks in games to sustain participant attention amid distractions in non-laboratory environments. The application for production experiments saves uncompressed audio data as WAV files, one recording per trial. Amplitude envelope features are used to isolate a speech token in each recording. This token is compared to the remainder of the recording to assess signal-to-noise ratio. Performing this analysis after each trial, we can supply the participant with feedback to improve recording quality. We tested the efficacy of using SpeechCollectR, collecting 1344 recordings of three types of affective prosody from 32 participants. 83% of these recordings were usable for analysis of prosodic variables. We present a bootstrap analysis demonstrating reliability of affective prosodic features despite variations in recording equipment and environment.