healthnexus tab ethics d4d

Datasheet for Dataset - Human Readable Format

🎯

Motivation

Why was the dataset created?

  • ID
    purpose-1
    Name
    Dataset purpose
    Response
    Enable AI research and critical insights into the use of voice as a biomarker of health via an ethically sourced, diverse, multi-institutional dataset linked to clinical information.
GrantorGrant NameGrant Number
National Institutes of Health (NIH)Bridge2AI: Voice as a Biomarker of Health - Building an ethically sourced, bioaccoustic database to understand disease like never before3OT2OD032720-01S1
📊

Composition

What do the instances represent?

CountsData TypeIDInstance TypeNameRepresentation
12523Spectrograms (513xN), acoustic/phonetic/prosodic features; raw waveforms not included in v1.0.instance-recordingsparticipants, sessions, and recordingsDerived recording instancesDerived voice data elements per recording (e.g., spectrogram matrices, static features) linked to se...
306Tabular phenotype data with one row per participant; associated data dictionary.instance-participantsparticipantsParticipant instancesAdult participants with demographic, clinical, and validated questionnaire responses.
  1. ID
    subp-1
    Name
    Disease cohorts
    Identification
    • Voice disorders
    • Neurological and neurodegenerative disorders
    • Mood and psychiatric disorders
    • Respiratory disorders
    • Pediatric voice and speech disorders (adult cohort only in v1.0)
    Distribution
    • Adult cohort only; 306 participants across five sites in North America.
  • ID
    distfmt-1
    Name
    Distribution formats and files
    Description
    • spectrograms.parquet (Parquet; spectrogram matrices and metadata)
    • phenotype.tsv (tab-delimited; one row per participant)
    • phenotype.json (data dictionary for phenotype)
    • static_features.tsv (tab-delimited; one row per recording with features)
    • static_features.json (data dictionary for features)
  • ID
    distdate-1
    Name
    Initial release date
    Description
    • 2024-11-27 (v1.0 first release)
  • ID
    access-1
    Name
    Access modality
    Description
    • Restricted (Credentialed Access) via Health Data Nexus; DUA and training required; no public download URLs provided for files.
🔍

Collection Process

How was the data acquired?

doi:10.57764/qb6h-em84
Bridge2AI-Voice v1.0
Bridge2AI-Voice: An ethically-sourced, diverse voice dataset linked to health information
Bridge2AI-Voice is a comprehensive, ethically sourced dataset enabling research on the human voice as a biomarker of health. The v1.0 release provides 12,523 recordings from 306 adult participants collected across five North American sites, with data derived from voice recordings (e.g., spectrograms, acoustic/phonetic/prosodic features) and linked clinical, demographic, and validated questionnaire information. Raw audio waveforms and free speech transcripts are not included in v1.0; only low-risk derivations are provided.
en
https://docs.b2ai-voice.org
2024-11-27
  • voice
  • audio
  • Bridge2AI
  • biomarker
  • clinical
  • spectrograms
  • credentialed access
  • multi-institutional
  • ID
    gap-1
    Name
    Gap addressed
    Response
    Addresses the lack of large, high-quality, diverse, and standardized multi-institutional voice datasets linked to other health biomarkers for robust AI research.
RoleNameORCIDAffiliation
Principal InvestigatorAlistair Johnsonperson-alistair-johnson-
Principal InvestigatorJean-Christophe Bélisle-Piponperson-jean-christophe-belisle-pipon-
Principal InvestigatorDavid Dorrperson-david-dorr-
Principal InvestigatorSatrajit Ghoshperson-satrajit-ghosh-
Principal InvestigatorPhilip Payneperson-philip-payne-
Principal InvestigatorMaria Powellperson-maria-powell-
Principal InvestigatorAnaïs Rameauperson-anais-rameau-
Principal InvestigatorVardit Ravitskyperson-vardit-ravitsky-
Principal InvestigatorAlexandros Sigarasperson-alexandros-sigaras-
Principal InvestigatorOlivier Elementoperson-olivier-elemento-
Principal InvestigatorYael Bensoussanperson-yael-bensoussan-
  • Bridge2AI-Voice Team
  1. ID
    sampling-1
    Name
    Clinical cohort sampling
    Is Sample
    • True
    Is Random
    • False
    Source Data
    • Patients at specialty clinics across five North American sites
    Is Representative
    • No (targeted cohorts by condition)
    Why Not Representative
    • Participants selected based on five predetermined disease groups (voice, neurological/neurodegenerative, mood/psychiatric, respiratory, pediatric)
    Strategies
    • Targeted clinical cohort sampling based on known voice-related conditions
  • ID
    rel-1
    Name
    Instance relationships
    Description
    • Recordings are nested within sessions and participants; features and spectrograms link via participant_id, session_id, and task_name.
  • ID
    confidential-1
    Name
    Clinical and questionnaire data
    Description
    • Contains clinical, demographic, and questionnaire information; released in de-identified, low-risk form.
  • ID
    sensitive-1
    Name
    Sensitive health-related data
    Description
    • Health, demographic, and questionnaire data linked to voice recordings (de-identified).
ID
deid-1
Name
De-identification
Description
  • HIPAA Safe Harbor identifiers removed (e.g., names, detailed geography, dates below year, contact and ID numbers, biometric identifiers).
  • State and province removed; country of data collection retained.
  • Transcripts of free speech audio removed prior to release.
  • Audio waveforms omitted from v1.0; only derived spectrograms and features released.
  1. ID
    acq-1
    Name
    Data acquisition
    Description
    • Standardized protocol including demographics, health and targeted questionnaires, disease-specific information, and voice tasks (e.g., sustained vowel).
    • Data captured via custom tablet application with headset when possible; consent obtained prior to collection.
    Was Directly Observed
    Yes (voice recordings, tasks)
    Was Reported By Subjects
    Yes (validated questionnaires)
    Was Inferred Derived
    Yes (features, spectrograms, ASR transcriptions)
    Was Validated Verified
    Standardized protocol and validated questionnaires; processing pipeline described and open-sourced.
  • ID
    mech-1
    Name
    Collection mechanisms
    Description
    • Custom tablet app; headset-based recording when possible; data exported from REDCap and converted using an open-source library (b2aiprep).
  • ID
    collectors-1
    Name
    Data collectors
    Description
    • Project investigators at specialty clinics screened patients and obtained consent; most participants completed a single session, some multiple sessions.
  • ID
    irb-1
    Name
    Ethical review
    Description
    • Data collection and sharing approved by University of South Florida Institutional Review Board; submitted for review to University of Toronto Research Ethics Board.
DescriptionIDNameUsed Software
Raw audio converted to mono and resampled to 16 kHz with a Butterworth anti-aliasing filter; STFT spectrograms computed with 25 ms window, 10 ms hop, 512-point FFT.prep-1Audio standardization and spectrogram extraction{'id': 'sw-torchaudio', 'name': 'Torchaudio'}
Temporal and acoustic characteristics extracted using OpenSMILE.prep-2Acoustic feature extraction{'id': 'sw-opensmile', 'name': 'OpenSMILE'}
Fundamental frequency, formants, and voice quality computed using Parselmouth and Praat.prep-3Phonetic and prosodic feature extraction{'id': 'sw-parselmouth', 'name': 'Parselmouth'}, {'id': 'sw-praat', 'name': 'Praat'}
Automatic speech transcriptions generated using OpenAI Whisper Large (free speech transcripts removed prior to release).prep-4Transcription{'id': 'sw-whisper', 'name': 'OpenAI Whisper Large'}
  • ID
    clean-1
    Name
    Data integration and standardization
    Description
    • Source data exported from REDCap and merged into phenotype and feature files; accompanying JSON data dictionaries provide variable descriptions; processing code available in b2aiprep.
  1. ID
    label-1
    Name
    Automatic transcription
    Description
    • ASR transcriptions generated using OpenAI Whisper Large; free speech transcripts removed from released data.
    Used Software
    • ID
      sw-whisper
      Name
      OpenAI Whisper Large
  • ID
    raw-1
    Name
    Raw audio waveforms
    Description
    • Raw audio collected but omitted from v1.0 release; future releases aim to include voice data with additional security precautions.
  1. ID
    ext-1
    Name
    External resources and documentation
    External Resources
    • Documentation Site
      https://docs.b2ai-voice.org
    • REDCap resource record (Zenodo citation provided in references)
    • b2aiprep open-source processing library
    Future Guarantees
    • Versioned DOIs available for releases.
    Archival
    • DOIs provided (versioned and latest).
    Restrictions
    • Access via Health Data Nexus under registered access with DUA and required training.
DescriptionDialectFormatIDMedia TypeNamePathTitle
Parquet dataset with participant_id, session_id, task_name, and 513xN spectrogram matrices derived f...subset-spectrograms-parquetapplication/x-parquetspectrograms.parquetspectrograms.parquetSpectrograms derived from raw audio
Tab-delimited table with demographics, acoustic confounders, and validated questionnaire responses (...delimiter:
header: true		subset-phenotype-tsvtext/tab-separated-valuesphenotype.tsvphenotype.tsvParticipant phenotype data
JSON data dictionary providing descriptions of columns in phenotype.tsv.JSONsubset-phenotype-jsonapplication/jsonphenotype.jsonphenotype.jsonPhenotype data dictionary
Tab-delimited table with one row per recording containing features derived from openSMILE, Praat, Pa...delimiter:
header: true		subset-static-features-tsvtext/tab-separated-valuesstatic_features.tsvstatic_features.tsvStatic acoustic features per recording
JSON data dictionary providing feature descriptions for static_features.tsv.JSONsubset-static-features-jsonapplication/jsonstatic_features.jsonstatic_features.jsonStatic features data dictionary
🚀

Uses

What (other) tasks could the dataset be used for?

  • ID
    task-1
    Name
    Intended tasks
    Response
    Research and development of AI methods for disease-related voice/speech changes, feature discovery, and health prediction using voice-derived representations linked to clinical data.
ID
license-1
Name
Access, license, and use terms
Description
  • License
    Bridge2AI Voice Registered Access License.
  • Access Policy
    Only credentialed users who sign the Data Use Agreement (DUA) can access the files.
  • Data Use Agreement
    Bridge2AI Voice Registered Access Agreement.
  • Required Training
    TCPS 2: CORE 2022.
  • ID
    future-impact-1
    Name
    Potential impacts on future use
    Description
    • Absence of raw audio in v1.0 may limit certain signal processing and modeling tasks; inclusion planned in future releases with additional safeguards.
📤

Distribution

How will the dataset be distributed?

doi:10.57764/qb6h-em84
ID
versioning-1
Name
Versioning and access
Description
🔄

Maintenance

How will the dataset be maintained?

1.0
ID
update-1
Name
Update plan
Description
  • Future releases aim to include voice waveforms with additional precautions to ensure data security; v1.0 provides low-risk derived data only.
Generated on 2025-11-09 10:17:34 using Bridge2AI Data Sheets Schema