InterSpeech 2021

How reliable are phonetic data collected remotely? Comparison of recording devices and environments on acoustic measurements
(3 minutes introduction)

Chunyu Ge (CUHK, China), Yixuan Xiong (CUHK, China), Peggy Mok (CUHK, China)
The COVID-19 pandemic posed an unprecedented challenge to phonetic research. On-site collection of speech data is difficult, if not impossible. The advancement of technology in mobile devices and online conference platforms offers the opportunity to collect data remotely. This paper aims to answer the question of how reliable speech data collected remotely are based on controlled speech. Seven devices, including smartphones and laptops, were used to record speech simultaneously, locally or on the cloud using ZOOM, both in a sound-attenuated lab and a conference room. Common acoustic measurements were made on these recordings. Local recordings proved to be reliable in duration, but not for recordings made on the cloud. Different devices have comparable performances in F0 and F1. The values acquired by different devices differ a lot for F2 and higher formants, spectral moments, and voice quality measures. These differences can lead to erroneous interpretation of segmental and voice quality contrasts. The recordings made remotely by smartphones and locally using ZOOM can be useful in studying prosody, but should be used with care for segments.