How do speech biomarkers work with an eCOA platform?
Speech is recorded as one more task in the participant’s existing eCOA app or on the site tablet, or in a dedicated recording app where the study needs one. Each recording and its metadata go to the analysis provider, and the readings come back into the same eCOA or EDC data flow as the trial’s other assessments.
Explainer for clinical operations, data management and digital endpoint teams, by Cephalgo · Updated November 2026
From the recording to the reading in four steps
The eCOA keeps the schedule and the participant’s record. The analysis provider turns audio into readings. Each hand‑over between the two is set in advance and tested before first participant in.
Record
The participant, or site staff at a visit, runs a speech task such as reading a passage or describing a picture. The app saves the audio with a timestamp and the device details.
Transfer
The audio file and its metadata go from the app or the eCOA server to the analysis provider by API or SFTP, under a pseudonymised participant ID, through a pre‑specified, validated transfer.
Analyse
The provider checks recording quality, computes the speech measures with an analysis version fixed for the study, and logs that version on every reading.
Return
Readings, quality flags and version details go back into the eCOA or EDC, or to data management as a transfer file, joined to the participant and visit.
The FDA’s guidance on digital health technologies asks that data and all relevant associated metadata be securely transferred to and kept in a durable electronic data repository, and treats the first such repository as the source data for inspection (FDA, 2023).
Three ways speech fits beside an eCOA
The choice depends on where the speech is recorded, at home or at a site visit, and on what the study’s eCOA can already run.
Recording inside the eCOA app
The eCOA provider builds the speech task into its app, on the participant’s own phone, a provisioned phone or a site tablet, and sends each recording to the analysis provider. Participants and sites keep one schedule, one login and one data flow. The eCOA provider controls the recording settings, so the audio specification belongs in its build requirements.
A separate recording app beside the eCOA
A dedicated app records speech while the eCOA runs questionnaires. It suits studies with no eCOA, or tasks the eCOA cannot present. It adds a second app for participants and a second system to validate, and the two data streams are joined on participant ID and visit.
Recorded visit interviews uploaded from site
Sites record rating interviews or speech tasks at the visit and upload the files to a secure portal. The same route serves recordings a trial already holds, which can be read again where consent allows. Site staff enter the participant ID and visit, so upload checks matter as much as audio quality.
The EMA guideline on computerised systems covers both eCOA and bring‑your‑own‑device set‑ups. It asks that data collected by any of these technologies be transferred to a server, made available to the investigator, and that every part of the process be described and validated (EMA, 2023, Annex 5).
What to agree before first participant in
The sponsor, the eCOA provider and the speech provider settle these points in the eCOA requirements, the data transfer agreement and the validation plan.
Audio format
Lossless audio, usually uncompressed PCM WAV, with the sample rate, bit depth and channel count written into the specification. Lossy compression, automatic gain and noise suppression change the signal that acoustic features are computed from, so the specification also states which processing the app turns off. Device changes ›
Metadata
Device model, OS build, app version, input route, task and language, recording start time with time zone, and transfer time. The EMA guideline asks that device data carry metadata such as device version, identifiers, firmware version, data originator and event timestamps, and that ePRO timestamps record the time of entry, not only of transmission.
Participant ID pseudonymisation
Only the study participant ID travels with the audio; the key stays with the site. Under GDPR Article 4(5), pseudonymised data can no longer be attributed to a person without additional information kept separately. The voice itself still identifies: HIPAA’s de‑identification rule lists voice prints among biometric identifiers (45 CFR 164.514).
Transfer
An API over HTTPS for each recording as it arrives, or scheduled SFTP batches, encrypted in transit, with a checksum per file and a routine for failed or repeated transfers. The EMA guideline asks that every transfer be pre‑specified and validated, and FDA’s 2024 questions and answers on electronic systems ask that the date and time of transfer from the device be in the audit trail.
Data standards
The data transfer agreement names the variables, format and frequency of returned readings. For submission, FDA’s digital health technology guidance asks sponsors to agree with the review division which data to submit and points to the FDA Data Standards Catalog. In CDISC SDTM, objective task‑based measures sit in the Functional Tests (FT) domain, questionnaires in QS, and clinical classifications in RS (CDISC QRS). A reading derived from speech is stored as its own variable, with its own test code and method, never as the scale item’s own score.
Audit trails: 21 CFR Part 11, ICH E6(R3) and the EMA guideline
Every system that creates or changes a record keeps a time‑stamped audit trail. FDA lists its components: date and time, the person making the change, and the old and new values, with the reason where applicable. The GCP rule for trial systems is ICH E6(R3): audit trails are not disabled, and initial entries, changes and deletions are documented with the reason where appropriate. In the EU, the EMA guideline sets out how this applies to trial systems. EudraLex Annex 11 is the computerised‑systems annex of EU good manufacturing practice.
Data protection: GDPR and HIPAA
Speech recorded to measure health is data concerning health, a special category under GDPR Article 9. Each party that processes it signs a data processing agreement, and transfers outside the EU need an adequacy decision or appropriate safeguards. US studies add HIPAA terms where a covered entity is involved. Vendor qualification ›
Validation of the integration and UAT
Each provider validates its own system; the sponsor runs user acceptance testing on the study build and keeps a record of each system’s validation status (ICH E6(R3)). The C‑Path consortia’s UAT recommendations include custom data transfers, and FDA applies validation to data transfers and interfaces between systems. For speech, test end to end: record on each supported phone, confirm the file arrives intact with its metadata, and check that the reading returns to the right participant and visit.
Who holds the blind
Readings are returned without arm information, and the analysis provider works blind to treatment. The EMA guideline restricts unblinded information to pre‑identified roles and keeps it out of audit trails that blinded users can see. Any unblinded view of readings sits with the independent statistician and data monitoring committee.
Own phones or study phones
eCOA runs on participants’ own phones, on phones provided for the study, or on site tablets. Each choice changes what the speech specification has to cover.
Participants’ own phones
The EMA guideline asks sponsors to set minimum technical specifications, to accept a range of models and operating systems, and to offer another way to take part to anyone who cannot or will not use their own device. The sponsor has no control over operating system updates on these phones, so device and OS metadata travel with every recording. BYOD ›
Study phones and site tablets
One model for everyone reduces variation between participants. The FDA guidance asks sponsors to name the models and versions used, and to record the timing and nature of updates during the trial. Study phones ›
Guidance, standards and best practice
The requirements on this page come from these documents.
U.S. Food and Drug Administration. Digital Health Technologies for Remote Data Acquisition in Clinical Investigations. Final guidance, December 2023.
Section IV.G: transfer of data and metadata to a durable electronic data repository, source data, and submission formats.
U.S. Food and Drug Administration. Electronic Systems, Electronic Records, and Electronic Signatures in Clinical Investigations: Questions and Answers. Guidance, October 2024.
Validation of data transfers and interfaces (Q7), audit trail components (Q12), and transfer from digital health technologies (Q22, Q23).
U.S. Food and Drug Administration. Part 11, Electronic Records; Electronic Signatures: Scope and Application. Guidance, August 2003.
How FDA applies 21 CFR Part 11, including its risk‑based approach to validation and audit trails.
European Medicines Agency, GCP Inspectors Working Group. Guideline on computerised systems and electronic data in clinical trials. EMA/INS/GCP/112288/2023, 9 March 2023.
Metadata, transfer, audit trails and blinding (sections 4 to 6 and Annexes 3 and 5), and eCOA and bring‑your‑own‑device systems (Annex 5).
International Council for Harmonisation. Guideline for Good Clinical Practice E6(R3). Final version, adopted 6 January 2025.
Sponsor duties for computerised systems, blinding in data governance (4.1), metadata and audit trails (4.2.2), and validation (4.3.4).
European Commission. EudraLex Volume 4, Annex 11: Computerised Systems. 2011.
The computerised‑systems annex of EU good manufacturing practice, including audit trails.
Gordon S, Crager J, Howry C, et al. Best Practice Recommendations: User Acceptance Testing for Systems Designed to Collect Clinical Outcome Assessment Data Electronically. Therapeutic Innovation & Regulatory Science. 2022;56(3):442–453. doi:10.1007/s43441-021-00363-z
The ePRO and PRO Consortia’s recommendations for sponsor UAT of eCOA systems, listed by the Critical Path Institute’s eCOA Consortium.
CDISC. Questionnaires, Ratings and Scales (QRS).
How questionnaires (QS) and functional tests (FT) are represented in SDTM.
U.S. Food and Drug Administration. Data Standards Catalog.
The study data standards FDA supports and requires for submissions.
Regulation (EU) 2016/679, General Data Protection Regulation. Articles 4(5), 4(15) and 9.
Definitions of pseudonymisation and data concerning health, and rules for special categories of personal data.
45 CFR 164.514, Other requirements relating to uses and disclosures of protected health information.
HIPAA de‑identification, with voice prints listed among biometric identifiers.
Speech biomarkers and eCOA: common questions
How do speech biomarkers work with an eCOA platform in a clinical trial?
Speech is recorded as one more task in the participant’s existing eCOA app or on the site tablet, or in a dedicated recording app where the study needs one. Each recording and its metadata go to the analysis provider, and the readings come back into the same eCOA or EDC data flow as the trial’s other assessments.
Can speech tasks run inside an existing eCOA app?
Yes. The eCOA provider builds the speech task into its app as one more task and sends each recording, with its metadata, to the analysis provider by API or SFTP. Participants keep one schedule and one login, and readings return into the eCOA or EDC. The audio specification belongs in the eCOA build requirements, because the eCOA app controls the recording settings.
What audio format should an eCOA app record for speech analysis?
Lossless audio, usually uncompressed PCM WAV, with the sample rate, bit depth and channel count fixed in the specification. Lossy compression, automatic gain and noise suppression change the signal that acoustic features are computed from, so the specification also states which processing the app turns off.
What metadata should travel with each speech recording?
Device model, OS build, app version, input route, task and language, recording start time with time zone, and transfer time, under a pseudonymised participant ID. The FDA’s 2023 guidance on digital health technologies asks that data and all relevant associated metadata be kept in a durable electronic data repository, and the EMA’s 2023 guideline lists device version, identifiers, firmware version, data originator and event timestamps.
Who validates a speech and eCOA integration?
Each provider validates its own system, and the sponsor or its designee runs user acceptance testing on the study build, including custom data transfers. FDA’s 2024 questions and answers on electronic systems apply validation to data transfers and interfaces between systems. For speech, the test runs end to end: record on each supported phone, confirm the file arrives intact with its metadata, and check that the reading returns to the right participant and visit.
Can participants record speech on their own phones in an eCOA study?
Yes. The EMA guideline on computerised systems allows bring‑your‑own‑device set‑ups when the sponsor sets minimum technical specifications, accepts a range of models and operating systems, and offers another way to take part to anyone who cannot or will not use their own phone. Because operating system updates on those phones are outside the sponsor’s control, device and OS metadata travel with every recording.
How Cephalgo works with your eCOA
Cephalgo reads speech for clinical trials from recordings you already hold, site‑visit recordings, and home sessions, where our speech tasks run inside your eCOA as one more performance outcome task. Your eCOA vendor sends each recording to our API. Readings on your scale, with quality flags, come back into the same system: eCOA or EDC, as CDISC ODM, SDTM‑aligned or your own format. The analysis version is pinned per study and logged on every reading. A phone call, a browser or Cephalgo’s app are used where a study has no eCOA. Capture platform ›Vendor qualification ›What is a speech biomarker ›