Fidelity of fricative measurements in remote data collection

This investigation explores whether differences between fricatives become compressed or shifted during remote recordings. Inputs for all test conditions (six devices across five teleconferencing platforms, two browsers, and internal and external microphones) were near-minimal English fricative-monophthong pairs read by a male speaker and female speaker. Differences of Mel frequency cepstral coefficients (MFCCs), spectral measures, and loudness between /s/ and each of /f, ʃ, ʒ/ under each test condition were compared to the original. MFCCs often showed expanded differences; other measures often showed compression. Even reversals in differences were observed. The findings highlight the need to carefully select and document remote recording conditions for fricative measures.