A voice feature means little until it sits next to the diagnosis, the visit, the consent and the clinical scale it is meant to track. Capture holds the study side of a speech biomarker project, with coded participant IDs, visit schedules, scales and an audit trail, and exports a clean dataset for your audio pipeline to join.
Free sandbox · No credit card · 21 CFR Part 11 aligned
| Assessment | Screen | Baseline | Wk 4 | Wk 8 | Wk 12 |
|---|---|---|---|---|---|
| Informed consent (incl. recording) | |||||
| Clinical rating scale (rater) | |||||
| Speech tasks recorded (external tool) | |||||
| Session log: device, room, task completed | |||||
| Participant questionnaire | D | ||||
| Adverse events and medications |
x = done at visit, D = diary. Demo schedule, not a real protocol.
What to know before you set up a speech biomarker study
The study type
Speech and voice biomarker research asks whether features of how a person speaks, such as speech rate, pause patterns, pitch variability, articulation or word choice, track a condition or its progression. The conditions people study this way include neurodegenerative disease, depression and mood disorders, respiratory disease and cognitive decline, because speech draws on motor control, cognition and emotion at once. The projects range from a single-site feasibility study with a few dozen people to multi-site validation studies that run alongside a conventional clinical trial.
Whatever the scale, the operational problem is the same. Audio files arrive from a recording app, a call platform or a clinic device. Feature extraction runs in a separate analysis environment. Meanwhile the clinical team runs visits, collects rating scales, records medication changes and adverse events, and keeps consent. A good setup keeps each of those in the system built for it and joins them on one stable, coded participant identifier. That join is where an EDC earns its place.
This is the same pattern as other digital endpoints, where sensor or device data is validated against a clinical reference. The principles are covered in digital biomarker validation study software and, for continuous device data, EDC for wearable device studies. For the underlying term, see biomarker in the glossary.
The link that matters
Each participant gets a coded study ID at enrolment. That is the identifier your recording tool and analysis pipeline use, so no name or contact detail travels with an audio file. Site coordinators see names; researchers and analysts see coded IDs. The exported dataset carries the same code on every row, with an automatic data dictionary, so joining features to clinical data is a keyed merge rather than a matching exercise.
Site coordinator
Jane Peterson
DOB: 04-Mar-1978
Subject 01-004
Sees direct identifiersResearcher
Subject 01-004
DOB: withheld
Coded ID only
Sees coded ID onlySame record, two roles. Row-level security enforces the split.
Session metadata
These fields are a starting point. They sit in a short eCRF completed by the person running the session, or in a participant questionnaire when sessions happen at home.
| Field | Why it matters | Typical form |
|---|---|---|
| Session date and time | Links the sample to the visit window and to time of day, which affects voice | Date and time |
| Task completed | Reading, sustained vowel, picture description, free speech and so on; partial completion is common | Multiple choice per task |
| Recording device and app version | Feature values shift with hardware and software changes | Dropdown plus text |
| Environment | Quiet room, background noise, clinic or home | Single choice |
| Participant state | Cold, hoarseness, fatigue, recent medication, caffeine | Single choice plus free text |
| Supervised or self-administered | Needed to compare in-clinic and remote sessions | Single choice |
| Quality flag from the recording tool | Lets the analysis team exclude on a documented rule, not on memory | Single choice |
| External file reference | Pointer to the recording in your secure audio store, using the coded ID only | Text |
| Reason session not done | Separates missing at random from missing because of disease progression | Dropdown |
Plan to keep audio in your own secure store. Keep a coded reference, never a name or contact detail, in the file name or this field.
Create a session metadata form, enter demo participants and download the coded dataset in the free sandbox.
The reference standard
A speech marker is only as convincing as the clinical measure it is compared against. A Parkinson's study will want the rater-completed motor examination alongside the recordings. A depression study will want a clinician-rated or self-reported severity scale. A cognitive study will want a validated test. These scales are collected on the same visit as the speech tasks, and the analysis compares the two at matched time points.
Capture lets you build those as site-completed eCRFs and participant questionnaires in the same study, on the same visit schedule, with skip logic, edit checks and calculated fields where the protocol needs them. Do check licensing before you reproduce a named instrument: Capture records licence details for participant questionnaires only, does not verify licences with rights holders and does not supply instrument wording, so the licence for site-completed scales stays with your study documents. Lists of ready-made forms are on the templates page, and how to capture patient-reported outcomes digitally covers the participant side.
Speech changes through the day, so many protocols fix a time window for recordings. Visit windows and scheduled tasks in the visit schedule make deviations visible. Record the actual time on the session form so the analysis can adjust for it, and treat out-of-window sessions as a deviation to be reviewed rather than deleted.
Remote speech collection lets you sample more often and reach people who cannot travel, which is the main reason this field is growing. The trade-off is less control of the environment. Participants can complete questionnaires and diaries in a phone browser without installing an app (see ePRO without app development), so a short pre-session checklist covering room, device and symptoms can be asked at the time of the recording.
Data quality
Edit checks can raise an automatic query when a session form breaks a rule, for example a task marked complete with no device recorded or a session outside its visit window. The monitoring dashboard and per-site filtering show which sites produce incomplete sessions, so retraining happens in week three rather than at database lock.
Sessions scheduled
240
Completed with all tasks
86%
Flagged for noise
9%
Two devices changed
What you do in Capture
Create visits and windows, or upload the protocol and let the AI study builder draft visits and forms for review. Nothing is saved without a human check.
Build the reference scales as eCRFs or questionnaires, plus a short session metadata form for device, environment, tasks and quality flag.
Site coordinators see names for scheduling. Researchers and analysts work from coded IDs only.
Enter demo participants, trigger edit checks, sign forms and download the export your analysis team will use.
Track completeness by site, review queries and export the coded dataset on a schedule.
Before the first recording
Recordings described in plain language, with storage, sharing and reuse stated, reviewed by your data protection officer.
A secure store outside the EDC, with access limited and file names containing coded IDs only.
One identifier used by the EDC, the recording tool and the analysis pipeline.
Device, environment, tasks and quality flag defined and tested.
Chosen, licensed where required, and scheduled on the same visits.
Reasons for failed sessions coded, not left as free text.
Columns, codebook and join key confirmed in the sandbox.
Where Capture stops
Be clear about the boundary. Capture is positioned as the study record, not as a recording app, an audio archive or a speech analysis tool, and it does not extract features. If your protocol needs a validated recording and feature extraction workflow, that comes from a specialist tool or your own pipeline. What Capture adds is the regulated study record around it: consent, visits, scales, adverse events, queries and an audit trail with the controls described as 21 CFR Part 11-aligned. For other digital health work, see EDC for digital health clinical trials and the FDA's digital health technologies material in this guide.
If a pilot study later becomes a pivotal or multi-site trial, the same study build scales up. Phase 1 through Phase 3 are all normal use cases, and the free sandbox lets you prove the data flow before committing.
Plan for audio to live outside Capture. Capture is built to hold the study record around the recordings, and audio capture or analysis is not something this page claims; ask the team if you need it. The study record covers coded participant IDs, visits, clinical scales, session metadata, consent and the audit trail. Audio stays in a separate secure store or recording tool, referenced by the coded ID.
Use one coded participant identifier across the recording tool, the analysis pipeline and the EDC. Capture exports a dataset with that code on each row and an automatic data dictionary, so your analysts join features to scale scores and visits with a keyed merge.
Yes, a voice recording of an identifiable person is personal data, and some uses may fall under biometric data rules. The exact classification depends on how the recording is used and where you run the study, so settle the legal basis and the consent wording with your data protection officer before collecting.
Capture can collect the questionnaires, diaries and session checklists a home session needs in a phone browser with no app install. The recording itself is taken with your own speech tool, and the session form records the device, environment and task completed.
You can build site-completed scales as eCRFs and participant-completed scales as questionnaires on the same visit schedule. Check licensing for any named instrument yourself, because Capture does not supply licensed wording or verify licences with rights holders.
No. Capture is an EDC and study platform, not a validated speech measurement. Validation of the speech feature as an endpoint is a scientific and regulatory question for your study team.
Yes. The free sandbox has every feature, with no credit card and no time limit. Build the session form and the visit schedule, enter demo participants and test the export. You pay only when you go live with real participants.
Keep exploring
Digital biomarker validation study software
Reference measures and device metrics.
EDC for wearable device studies
Continuous device data.
EDC for digital health clinical trials
Digital endpoints in trials.
EDC for mobile health studies
Phone-based data collection.
Role-based access control
Who sees names and who sees codes.
Data pseudonymization in EDC
How coded IDs and PII segregation work.
GDPR compliant clinical trial software
EU data protection basics.
Free sandbox with every feature. No credit card, and you pay only when you go live with real participants.