Related Experiment Video
Updated: Sep 18, 2025
Foreign Accent and Forensic Speaker Identification in Voice Lineups: The Influence of Acoustic Features Based on Prosody
Published on: September 27, 2024
Comparing Phoneme Speech Recordings and Acoustic App Data Capture Experience for Android and iOS Mobile Device Users
Kara Chappie1, Stefan Kell1, Donnie Qi2
1Pfizer Research & Development, Pfizer, Inc., Cambridge, Massachusetts.
Objective:
Growth in telehealth and interest in decentralized clinical trials have motivated the need to understand Android and iPhone Operating System (iOS) impacts on remote, mobile app speech capture. This research investigated Android and iOS device differences in 16 acoustic, physiologically based speech features extracted from phonemes /i/ and /m/, how in-app instructions impacted maximum phonation time (MPT) for /a/, and a mobile app design consideration to capture quality signals for acoustic analyses.
Methods:
Acoustic features were auto-extracted from vocalizations recorded on the personal cell phones of 6505 subjects as part of a 6-8 weeks longitudinal, at-home, clinical trial. Feature averages were computed for 693 self-reported healthy participants (no chronic or acute conditions). Wilcoxon two-sample tests comparing mean feature values from Android and iOS speech recordings were computed for these self-reported healthy participants.
Results:
Periodic measures such as harmonicity differed between Android and iOS, with iOS registering more periodic content. Energy-related features demonstrated lower levels of high-frequency energy in the iOS results. Signal-to-noise ratio and coefficient of variation in fundamental frequency measured similarly across Android and iOS, as did the first three formants for /i/. All features showed more variability in Android devices than in iOS devices. Average background noise intensity levels were lower for Android. MPT averages were longer on the sustained /a/ task after a study pause period.
Conclusions:
Measurement differences were found between Android and iOS devices for several features that have historically been used to describe disease change. Device differences impacted participant experience in recording their speech, with iOS users having more difficulty passing in-app intensity-based background noise checks. An app instruction update to the sustained /a/ task during the study pause period resulted in longer MPT averages, demonstrating the importance that app instructions play in remote clinical trials.
Trial Registration Number:
NCT04748445.
