Mobile device audio processing
When a mobile device’s audio processing changes
Before a recording reaches analysis, the device can process it: gain control, compression, noise suppression, codecs. Over a long study that processing can change, and a speech measure can move with it.
The problem
A processing change can look like disease change
When the audio processing changes, a single measure jumps. The multi-stream reading limits the drift.
True progressionA single speech measureCephalgo multi-stream readingProcessing change
Fixed device or bring your own device (BYOD)
What to check in each set-up
Over a long study, each set-up meets processing change in different places. Each comes with a question to ask.
Updates
A lock delays updates. Some Android phones allow a firmware build to be pinned; iPhone allows deferral only.
AskIs the firmware build pinned, and recorded with each file?
Google Play system updates
They install at the next reboot and can change software codecs.
AskIs the audio stored uncompressed?
A replacement phone
The same model can arrive with a newer build or a different chip variant.
AskHow is a replacement phone treated in the analysis?
App releases
The recording app decides the audio source, format and in-app processing.
AskIs the app version logged with every file?
End of support
Many Android models have a published security-support window, counted from launch.
AskDoes support run past the last planned visit?
Microphone and case
Dust and case membranes can change what the microphone picks up.
AskIs the case fixed, and is a per-participant baseline kept?
Updates
Operating system and browser updates arrive on the participant's schedule.
AskAre operating system and app versions logged with every recording?
Many phone models
Model differences are measured on real devices and recalibrated from the audio.
AskHow are model differences handled?
Unprocessed capture
The recording route is checked at enrolment and logged with every file.
AskWhich recording mode is checked at enrolment, and logged?
Participant settings
On iPhone the participant can choose a microphone mode in apps that support it, and any phone can take a headset.
AskAre the microphone mode and input route logged?
A new phone
A replacement phone is found from the audio and recalibrated.
AskHow is a step change found and corrected?
In both: Cephalgo logs the recording mode, file format and versions in force, and measures the processing on real devices.
Recalibration
Recalibrated from the recordings themselves
Recalibration runs from the audio alone; device and version logs, where kept, add a second check. So participants record on their own phones, and fixed devices are covered too.
Step 1
Find the processing periods
A recording fingerprint assigns each recording to the period in which one processing setting was in force, from the audio alone.
Step 2
Anchor and re-estimate
The components of the multi-stream reading that are most stable across processing states anchor a re-estimation of each measure across periods.
Step 3
Your analysis, unchanged
The model and the study's measure definitions stay fixed; only the shift tied to each processing period is re-estimated, and the longitudinal analysis runs as planned.
Bring your own device
Built for participants' own phones
Participants record on their own phone or tablet. Changes in their processing are recalibrated from the recordings.
Fixed device
Logged per session
Device, operating-system build, app and analysis version are logged per session, including for replacement units.
What regulators expect
Keep a record of every update
“Sponsors should keep a record of the timing and nature of any updates for each DHT.” And where the measurement changes, “sponsors should compare data from DHTs before and after the update.”FDA, Digital Health Technologies for Remote Data Acquisition in Clinical Investigations, December 2023, section IV.H.4
What Cephalgo does
Limits drift, and logged at every session
The reading
Three streams
The streams respond differently to the recording chain, which limits drift in the multi-stream reading when one of them moves.
The recording
Uncompressed audio
Accepted recordings are kept as uncompressed audio, which limits exposure to codec updates.
The record
Versions per session
Device, operating-system build, app and analysis version are logged with every file, so every update gets a before-and-after comparison.
Presented
The work behind it
Robustness to on-device processing change, presented in 2026.
Basel, June 2026
Digital Biomarkers in Clinical Trials Summit
When the Device Changes Mid-Trial: Evaluating Speech-Based Digital Endpoint Robustness to On-Device Preprocessing Variability in Longitudinal Remote Monitoring
Request the poster
Philadelphia, September 2026
ISCTM Autumn Conference
When the Device Changes Mid-Trial: Speech-Based Digital Endpoint Robustness to On-Device Preprocessing Variability
Request the poster
Contact
Planning a fixed-device or BYOD study?
Bring the device plan, and we will go through each point for your protocol.
Get in touch →