Examining Recording Quality from Two Methods of Remote Data Collection in a Study of Vowel Reduction

Remote recording quality of speech data varies significantly by recording devices, formats, and platforms, and past work has asserted that fine-grained sociolinguistic work should not be conducted remotely, while broad questions, such as analyses of the relative position of phonemes in the vowel space, may be amendable to remote data collection. In this study, lossless offline remote recordings taken via smartphone and lossy web-based recording performed over Gorilla are compared to traditional laboratory recordings in order to determine how accurately the remote options replicate a study of English vowel reduction. Four measures of reduction are examined: Relative duration, Euclidean distance, Pillai scores, and normalized formant values of stressed and unstressed vowels. Temporal analyses and Pillai scores were unaffected by recording method, while Euclidean distance and formant values exhibited some statistically significant changes but remained largely in line with laboratory data. These findings indicate that remote offline recording via smartphone or Gorilla may hold promise for studying vowel reduction and other phenomena requiring a similar degree of precision in formant analysis, but researchers should be aware of the specific distortions likely to be incurred with each method, with smartphone recordings having a stronger impact than Gorilla on low and back vowels.

Automated summary

This study compares remote recordings taken via smartphone and web-based recording through Gorilla with traditional laboratory recordings. It finds that remote offline recording can accurately replicate studies of English vowel reduction, but researchers should be aware of specific distortions that may occur with each method.