The PROCESS-2 dataset contains human speech recordings and associated clinical metadata collected for research into cognitive assessment and digital health technologies. Access is granted solely for legitimate scientific research purposes.

2. Definitions

Dataset refers to all audio recordings, transcripts, metadata, derived features, and associated documentation comprising the PROCESS-2 release.
Recipient refers to the individual researcher and their affiliated institution requesting access.

3. Permitted Use

The Recipient agrees to use the Dataset only for:

non-commercial academic research,
methodological development,
educational purposes,
scientific publication.
Clinical deployment or commercial product development requires separate written permission.

4. Prohibited Uses

The Recipient shall NOT:

Attempt to identify or re-identify any participant.
Combine the Dataset with external information for identification purposes.
Redistribute, share, sublicense, or publicly host the Dataset.
Upload Dataset contents to public repositories or model training datasets.
Use the Dataset for surveillance, profiling, or discriminatory applications.
Use the Dataset for training biometric identification systems.

5. Data Security Requirements

The Recipient agrees to:

store data on secure institutional systems,
restrict access to authorised project personnel only,
prevent unauthorised copying or transfer,
delete local copies upon project completion or upon request.

6. Publication Requirements

Publications using the Dataset must:

cite the PROCESS-2 dataset publication,
acknowledge the dataset creators,
avoid inclusion of identifiable speech excerpts.

7. No Attempt at Re-Identification

The Recipient explicitly agrees:

No attempt will be made to determine the identity of any participant.
Violation results in immediate termination of access.

8. Redistribution Prohibition

The Dataset may not be redistributed in whole or in part without written permission from the data custodians.
Derived models must not permit reconstruction of identifiable speech data.

9. Ethical Compliance

The Recipient confirms compliance with:

institutional ethics policies,
applicable data protection regulations,
GDPR requirements where applicable.

10. Termination of Access

Access may be revoked at any time if terms are violated. Recipients must immediately delete all copies upon termination.

11. Disclaimer

The Dataset is provided "as is" without warranty. The creators accept no liability for misuse.

12. Agreement

By requesting or downloading the Dataset, the Recipient agrees to all terms stated in this agreement.

By clicking below, you confirm that you have read and understood the terms above. Please provide your professional details to request access. All fields are required.

PROCESS-2: Remote Speech Dataset for Cognitive Assessment

Dataset Summary

PROCESS-2 is a large-scale remote conversational speech dataset designed for research on speech-based biomarkers of cognitive decline. The dataset was collected using the CognoSpeak automated digital assessment platform and contains speech recordings from older adults spanning healthy cognition, mild cognitive impairment (MCI), and dementia.

The dataset captures ecologically valid speech produced in real-world environments using participants’ personal devices, enabling the development and evaluation of machine learning models for automated cognitive assessment and digital health screening.

PROCESS-2 provides:

400 participants across three diagnostic groups
1,200 speech recordings
manually generated transcripts
demographic and clinical metadata
predefined train/test splits for benchmarking

The dataset supports research in speech processing, clinical AI, digital biomarkers, and computational neurolinguistics.

Dataset Structure

Each participant contributes recordings from three speech elicitation tasks:

Task	Description	Duration
Semantic Fluency Task (SFT)	Produce words from a semantic category	60 seconds
Phonemic Fluency Task (PFT)	Produce words beginning with a letter	60 seconds
Cookie Theft Description (CTD)	Describe a complex visual scene	Variable

Directory Layout

PROCESS-2/
 ├── PROCESS-2_rec__XXX/
 │     ├── PROCESS-2_rec__XXX__SFT.wav
 │     ├── PROCESS-2_rec__XXX__SFT.txt
 │     ├── PROCESS-2_rec__XXX__PFT.wav
 │     ├── PROCESS-2_rec__XXX__PFT.txt
 │     ├── PROCESS-2_rec__XXX__CTD.wav
 │     └── PROCESS-2_rec__XXX__CTD.txt

where XXX is an anonymised participant identifier.

Data Fields

Participant metadata are provided in metadata.csv.

Field	Description
dir_name	Participant identifier
diagnosis	Dementia / MCI / HC
age	Age at assessment
gender	Self-reported gender
MMSE	Mini-Mental State Examination score (if available)
Split	Train or Test partition

Data Collection

Acquisition Platform

Data were collected using the CognoSpeak browser-based digital cognitive assessment system.

Participants accessed the assessment remotely using personal laptops or tablets through Google Chrome. A virtual conversational agent delivered instructions and recorded audio responses automatically.

Recruitment and Diagnosis

Participants were recruited through:

UK National Health Service (NHS) memory services
primary and secondary care pathways
Join Dementia Research (JDR) registry
Great Minds research registry

Clinical diagnoses followed NICE dementia assessment guidelines and were assigned independently by qualified clinicians.

Diagnostic labels were determined prior to computational analysis, and clinicians were not exposed to machine learning outputs.

Inclusion Criteria

Referral to memory clinic with suspected cognitive impairment
Capacity to provide informed consent

Exclusion Criteria

Lack of consent capacity
Insufficient English comprehension
Severe speech impairment preventing participation
Severe motor impairment preventing interaction

Hearing impairment, stroke history, Parkinson’s disease dementia, and psychiatric comorbidity were not exclusion criteria unless affecting consent capacity.

Recording Conditions

Recordings were obtained in real-world environments including participants’ homes and clinical settings.

Some sessions involved assistance from clinicians or care partners; conversational contributions from assisting speakers were retained to preserve real-world deployment characteristics.

Participants completed a single assessment session. Follow-up sessions were excluded from this release.

Preprocessing

All audio files were:

converted to mono
resampled from 44.1 kHz to 16 kHz
loudness normalised to −23 LUFS (EBU R128)

Transcripts were manually generated and preserved without post-hoc correction.

Intended Uses

PROCESS-2 supports research in:

dementia detection
speech biomarker discovery
clinical machine learning
conversational AI
multimodal health assessment
ASR evaluation on clinical speech
longitudinal monitoring method development

Benchmarking

The dataset includes predefined train/test splits enabling reproducible benchmarking.

Baseline experiments and full analysis pipelines are available:

👉 https://github.com/Madhurananda/PROCESS-2

Ethical Considerations

The study received ethical approval from:

NRES Committee South West–Central Bristol REC Number: 16/LO/0737

All participants provided informed consent for research use and controlled data sharing.

The dataset is intended strictly for research purposes related to health, speech, and machine learning.

Limitations

Remote recordings introduce acoustic variability.
Manual transcripts vary slightly across annotators.
Medication usage data were not collected.
Dataset represents cross-sectional assessments rather than longitudinal monitoring.

Citation

If you use this dataset, please cite:

[PROCESS-2 dataset publication — add citation after acceptance]

Contact & Support

If you encounter any technical issues with the dataset, have questions regarding the clinical metadata, or need to report a potential data concern, please contact the PROCESS-2 team:

General Inquiries: process.challenge [at] gmail [dot] com

Please include "PROCESS-2 Dataset Query" in your email subject line for a faster response.

License

This dataset is released under the Data Use Agreement.

Access and Data Governance

This dataset contains sensitive human speech recordings and is distributed under controlled access.

Researchers must request access and agree to the PROCESS-2 Data Use Agreement prior to download.

The dataset must not be redistributed or used for participant identification.

Downloads last month: 4

Total file size:

2.45 GB