healthnexus tab citationModal d4d

Grantor	Grant Name	Grant Number
National Institutes of Health (NIH)	Bridge2AI: Voice as a Biomarker of Health - Building an ethically sourced, bioacoustic database to understand disease like never before	3OT2OD032720-01S1

📊

Composition

What do the instances represent?

Instances

Counts	Data Type	ID	Instance Type	Label	Missing Information	Name	Representation	Sampling Strategies
12523	Derived features from raw audio including spectrograms and acoustic/phonetic/prosodic features.	inst-recordings	Recordings (per session and task)	Task names per recording (e.g., sustained phonation vowel tasks); no raw audio included in v1.0.	{'id': 'miss-1', 'name': 'Removed elements', 'missing': ['Raw audio waveforms (omitted in v1.0)', 'Transcripts of free speech audio'], 'why_missing': ['Privacy and low-risk release; de-identification measures']}	Voice-derived recordings	Voice recordings-derived data (spectrograms and engineered features) aligned to session and task met...	{'id': 'samp-1', 'name': 'Cohort sampling', 'is_sample': [True], 'is_random': [False], 'source_data': ['Patients presenting at specialty clinics/institutions across five North American sites'], 'is_representative': ['not specified'], 'strategies': ['Targeted enrollment of five predetermined clinical groups (respiratory, voice, neurological, mood/psychiatric, pediatric)']}
306	Demographics, acoustic confounders, validated questionnaires, disease-specific information collected...	inst-participants	Participants (adult cohort in v1.0)	Disease cohort categories and clinical attributes captured in phenotype file.		Participants	Individual participants with linked demographic, clinical, and questionnaire data.

Subpopulations

Distribution	ID	Identification	Name
Not specified in this record	subpop-1	v1.0 includes only adult participants	Adult cohort
Per-group counts not specified in this record	subpop-2	Voice disorders, Neurological and neurodegenerative disorders, ... (+3 more)	Clinical condition groups

Distribution Formats

Description	ID	Name
Credentialed access via Health Data Nexus portal; files available after DUA and required training completion	dist-portal	Distribution channel
Parquet (spectrograms), TSV (phenotype and static features), JSON (data dictionaries)	dist-formats	File formats

Distribution Dates

ID
distdate-1
Name
Initial public release
Description
- 2024-11-27 (v1.0)

🔍

Collection Process

How was the data acquired?

ID *

b2ai-voice-v1.0

Name

Bridge2AI-Voice

Title

Bridge2AI-Voice: An ethically-sourced, diverse voice dataset linked to health information v1.0

Description

The Bridge2AI-Voice project seeks to create an ethically sourced flagship dataset to enable future research in artificial intelligence and support critical insights into the use of voice as a biomarker of health. Bridge2AI-Voice v1.0 provides 12,523 recordings for 306 participants collected across five sites in North America. Participants were selected based on known conditions which manifest within the voice waveform including voice disorders, neurological disorders, mood disorders, and respiratory disorders. The initial release contains data considered low risk, including derivations such as spectrograms (not the original voice recordings), as well as detailed demographic, clinical, and validated questionnaire data.

Page

https://doi.org/10.57764/qb6h-em84

Issued

2024-11-27

Created On

2024-11-27

Language

English

Keywords

voice
bridge2ai
audio
spectrograms
phenotype
parquet
tsv
health

Created By

Alistair Johnson
Jean-Christophe Bélisle-Pipon
David Dorr
Satrajit Ghosh
Philip Payne
Maria Powell
Anaïs Rameau
Vardit Ravitsky
Alexandros Sigaras
Olivier Elemento
Yael Bensoussan

Status

bibo:draft

Addressing Gaps

ID
gap-1
Name
Gaps addressed
Description
Addresses limitations of prior work with small datasets, limited demographic diversity, and non-standardized protocols.
Response
Provides a large, multi-institutional, standardized, ethically sourced dataset with diverse demographics and linked clinical information.

Acquisition Methods

ID
acq-1
Name
Data acquisition
Description
- Standardized protocol with demographic information, health questionnaires, targeted acoustic confounders, disease-specific information, and voice recording tasks (e.g., sustained vowel phonation)
- Data exported from REDCap and converted using an open-source library developed by the team
Was Directly Observed
yes (voice tasks recorded via headset/tablet app)
Was Reported By Subjects
yes (validated questionnaires and self-reported items)
Was Inferred Derived
yes (acoustic/phonetic/prosodic features and spectrograms derived from raw audio; ASR transcriptions generated then removed for free speech)
Was Validated Verified
yes (standardized multi-site protocol and IRB/REB oversight)

Collection Mechanisms

ID
mech-1
Name
Collection mechanisms
Description
- Custom tablet application for data capture; headset used when possible
- Data managed in REDCap; conversion and preprocessing via open-source b2aiprep library
- Standardized tasks including sustained phonation

Data Collectors

ID
collectors-1
Name
Data collectors
Description
- Project investigators at specialty clinics and institutions; participants screened for inclusion/exclusion prior to visit; consent obtained

Collection Timeframes

ID
timeframe-1
Name
Collection timeframe
Description
- Data collected across five sites in North America; specific collection dates not provided in this record; first public release on 2024-11-27

Ethical Reviews

Description	ID	Name
Data collection and sharing approved by the University of South Florida Institutional Review Board	ethics-usf	IRB approval (USF)
Submission to the University of Toronto Research Ethics Board for review	ethics-utoronto	REB submission (U of Toronto)

Preprocessing Strategies

ID

prep-1

Name

Audio preprocessing and feature derivation

Description

Raw audio converted to monaural and resampled to 16 kHz with a Butterworth anti-aliasing filter
Spectrograms computed via STFT (25 ms window, 10 ms hop, 512-point FFT)
Acoustic features extracted with OpenSMILE
Phonetic and prosodic features computed using Parselmouth and Praat (e.g., F0, formants, voice quality)
Transcriptions generated with OpenAI Whisper Large (free speech transcripts later removed from release)

Used Software

ID	Name	URL	Version
sw-opensmile	openSMILE	https://www.audeering.com/research/opensmile
sw-praat	Praat	https://www.fon.hum.uva.nl/praat/
sw-parselmouth	Parselmouth	https://parselmouth.readthedocs.io/
sw-torchaudio	torchaudio	https://pytorch.org/audio	2.1
sw-whisper	OpenAI Whisper (Large)	https://github.com/openai/whisper

Cleaning Strategies

ID
clean-1
Name
Data cleaning and packaging
Description
- Conversion/merging of source data into phenotype files and spectrogram parquet using b2aiprep
- Creation of data dictionaries (phenotype.json, static_features.json) describing each column/feature

Labeling Strategies

ID
label-1
Name
Labeling/transcription
Description
- ASR transcriptions generated using Whisper Large; transcripts of free speech were removed prior to release

Raw Sources

ID
raw-1
Name
Raw data availability
Description
- In v1.0, audio waveforms are omitted; only spectrograms and derived features are provided. Raw audio may be considered for future releases with additional safeguards.

External Resources

Archival	External Resources	ID	Name	Restrictions
DOI landing page provides versioned records (v1.0)	https://docs.b2ai-voice.org	ext-docs	Project documentation	Credentialed access with DUA and training required for files
	https://doi.org/10.5281/zenodo.14148755	ext-redcap	Bridge2AI Voice REDCap (v3.20.0) - Zenodo
	https://github.com/sensein/b2aiprep	ext-b2aiprep	b2aiprep library (open source)

Is Deidentified

ID

deid-1

Name

De-identification

Description

HIPAA Safe Harbor identifiers removed (e.g., names, detailed dates, contact info, SSNs, MRNs, device IDs, URLs, biometric identifiers)
State/province removed; country of data collection retained
Transcripts of free speech removed
Audio waveforms omitted in v1.0; only derived data released

Sensitive Elements

ID
sens-1
Name
Sensitive data considerations
Description
- Contains demographic, clinical, and questionnaire data; released elements considered low risk after de-identification

Maintainers

ID
maint-1
Name
Hosting and maintenance
Description
- Health Data Nexus
- Temerty Centre for AI Research and Education in Medicine (Supported by the Temerty Foundation)

Errata

Is Tabular

Mixed (Parquet, TSV, JSON)

Subsets

Description	Dialect	Encoding	Format	ID	Is Data Split	Is Subpopulation	Media Type	Name	Path	Title
Parquet file storing dense time-frequency representations derived from raw audio waveforms with part...		UTF-8		subset-spectrograms-parquet	False	Adult cohort (v1.0 only)	application/x-parquet	spectrograms.parquet	spectrograms.parquet	Spectrograms derived from raw audio
Tab-delimited file containing demographics, acoustic confounders, and responses to validated questio...	delimiter: header: True	UTF-8		subset-phenotype-tsv	False	Adult cohort (v1.0 only)	text/tab-separated-values	phenotype.tsv	phenotype.tsv	Phenotype data (participant-level)
JSON data dictionary detailing each phenotype column and its description.		UTF-8	JSON	subset-phenotype-json	False	Adult cohort (v1.0 only)	application/json	phenotype.json	phenotype.json	Data dictionary for phenotype data
Tab-delimited file with one row per recording; includes features from openSMILE, Praat, parselmouth,...	delimiter: header: True	UTF-8		subset-static-features-tsv	False	Adult cohort (v1.0 only)	text/tab-separated-values	static_features.tsv	static_features.tsv	Static audio features
JSON data dictionary describing each feature present in static_features.tsv.		UTF-8	JSON	subset-static-features-json	False	Adult cohort (v1.0 only)	application/json	static_features.json	static_features.json	Data dictionary for static features

Motivation

Composition

Collection Process

Uses

Distribution

Maintenance