--- title: "Healthcare AI Data Collection: Speech, Signals, Video" url: https://ypai.ai/solutions/healthcare/custom-data-collection/ description: "Custom healthcare AI data: clinical speech, cardiology signals, procedure video, wearables, biosignals. Nordic languages, per-contributor consent, EEA default." source: "src/copy/routes (route /solutions/healthcare/custom-data-collection/)" --- # The data your clinical model needs is rarely for sale. We collect it. > Custom healthcare AI data: clinical speech, cardiology signals, procedure video, wearables, biosignals. Nordic languages, per-contributor consent, EEA default. Consultation speech, dictation and command audio, cardiology signals, procedure video, wearables and biosignals. Nordic clinical language cannot be scraped at volume, so collection is the only source. ## Three questions decide whether your model ever sees clinical data. Clinical data cannot be scraped, bought casually, or borrowed from a benchmark. Whether your programme gets it comes down to consent, reach and custody. ### Does this data exist to buy? - Rarely. Real consultations, dictations, cardiology traces and procedure footage are seldom on a marketplace. Where a rights-cleared corpus exists we license it; where it does not, collection is the only source, and we collect only what is missing. ### Will the consent pass your DPO? - Every contributor consents to a stated purpose, under GDPR Article 7, with Article 9 handling where biometrics are involved. The consent record, the withdrawal path and the audit trail ship with the data. ### Can the cohort be reached? - Native-speaker contributors, in the languages and settings the model will see: the clinic, the home, the device chain. Quotas and QA are written before capture starts. You judge the files against that. ## What we capture for clinical AI teams. Four capture programmes we scope most often. Cohort, setting, devices, languages and QA are set per engagement. ### Clinical speech, dictation and commands - The encounters, dictations and spoken commands a clinical voice model must learn from. - Consultation-style dialogue, clinician dictation and command audio from native-speaker contributors in Danish, Norwegian, Swedish, Finnish and the other languages your deployment needs. Consent and rights controls configured for the engagement; diarisation and clinical-terminology transcription on request. - **Cohort**: Native speakers · languages to match - **Setting**: Consultation dialogue · dictation · commands - **Consent**: confirm. Article 7 · documented per contributor Consent and lineage on every file ### Cardiology signals - ECG and echocardiography signals for cardiology models. - ECG traces and echocardiography recordings captured or sourced where the rights setup holds, timestamp-aligned and delivered against your schema. Annotation by credentialed reviewers is available as a second engagement. - **Signal**: ECG · echo recordings - **Alignment**: Timestamped against the agreed schema - **Rights**: confirm. Checked before capture ### Procedure and endoscopy video - What the camera sees in a procedure, captured under the clinic's own rules. - Procedure and endoscopy video collected in clinical environments with ethics, rights and de-identification boundaries agreed before a single frame is captured. Segmentation and annotation live on the imaging route. - **Boundary**: Ethics and rights agreed first - **Capture**: In the clinical environment · to quota - **De-identification**: flag. Boundary stated per engagement Held to the agreed quota · logged ### Wearables and biosignals - The signals a remote-monitoring model is trained on, from consented contributors. - Wearable, IoT and environmental sensor streams for monitoring and anomaly-detection programmes: self-hosted ingestion plus consent-anchored contributor recruitment, calibrated and timestamp-aligned for training. - **Source**: Consented contributors · screened to the programme - **Ingestion**: Self-hosted · calibrated - **Consent**: confirm. Anchored per contributor ## One run, five stations. Green only where a person confirms. Every collection programme moves through the same stations. Each station is signed off and documented, and that documentation ships with the delivery. ### The specification, written with you - Cohort, setting, device chain, languages, quota and QA criteria in one document you sign off before capture starts. ### A person, consenting to a purpose - GDPR Article 7 consent per contributor, Article 9 handling where biometrics are involved, withdrawal path documented. ### The clinic, the home, the device chain - Captured in the agreed environments with the agreed devices, staged to quota, edge cases included by design. ### A reviewer rules on every batch - Agreed quality criteria applied per batch, exceptions surfaced, the ruling recorded. ### Files with their paperwork - Delivery against the schema with the consent record, the audit trail and lineage on every file. Scope, quotas and QA stages are set per engagement. ## If the data exists, you should not pay to collect it. ### License first, collect the gap. - We match your need against rights-cleared inventory, our own and partner-sourced. What exists is licensed, with traceability from dataset to consent and licence basis. Only the missing part is collected. ### Synthetic, where real data cannot go. - Where privacy, scarcity or consent constrains real data, synthetic clinical data is generated against a defined distribution and calibrated against real-data baselines, with the utility-versus-privacy trade-off stated. ## Collected in the clinic. Kept in the strictest room. A Norwegian company under GDPR. Clinical project data is stored in Europe by default and processed in the EEA by default, with the paperwork to prove both. US healthcare work is handled HIPAA-compliant. - **Jurisdiction**: Norway · GDPR-native - **Storage**: European by default - **Processing**: EEA by default - **Consent**: Documented per contributor - **US engagements**: HIPAA-compliant handling - **Contracts**: Standard DPA terms · SCCs available - **Erasure**: 30-day end-of-contract SLA - **Withdrawal**: Workflow with audit trail - [The healthcare hub](https://ypai.ai/solutions/healthcare/) - [Clinical AI systems](https://ypai.ai/solutions/healthcare/clinical-ai-systems/) - [Clinical model evaluation](https://ypai.ai/solutions/healthcare/clinical-model-evaluation/) - [Clinical text and coding](https://ypai.ai/solutions/healthcare/clinical-text-and-coding/) - [Medical imaging annotation](https://ypai.ai/solutions/healthcare/medical-imaging-annotation/) - Safeguards for US healthcare speech work - [De-identifying clinical audio](https://ypai.ai/blog/compliance/how-to-de-identify-audio-data-hipaa/) ## Start with one cohort, one setting, one modality. We scope against your model programme, write the specification with you and test the pipeline on a pilot batch before full capture. Scope · Specification · Pilot batch · Acceptance · Full capture A bounded pilot carries its own acceptance criteria. You judge the files against what was agreed. Briefs are treated as confidential. We are used to data that cannot leave the EEA and programmes that cannot be named. Tell us the model programme, the modality and the languages or signals you need. We reply with a feasibility read. We come back with a feasibility read on your programme and modality. Enquiry details are treated as confidential.