healthnexus tab files d4d

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

📊

Composition

What do the instances represent?

Instances

Counts	Data Type	Instance Type	Missing Information	Representation	Sampling Strategies
12523	Derived features (spectrograms, acoustic, phonetic/prosodic); no raw audio waveforms in v1.0	Audio-derived instance	{'missing': ['Original audio waveforms'], 'why_missing': ['Omitted in initial low-risk release; planned for future releases with additional safeguards']}, {'missing': ['Transcripts of free speech audio'], 'why_missing': ['Removed during de-identification']}	Voice-derived recordings (spectrograms/features) per recording	{'is_sample': ['yes'], 'is_random': ['no'], 'source_data': ['Patients presenting at specialty clinics across five North American sites'], 'is_representative': ['no'], 'why_not_representative': ['Participants were selected based on membership in predefined disease cohorts'], 'strategies': ['Targeted enrollment by predefined disease categories']}
306	Demographics, clinical, and validated questionnaire responses	Participant		Participant-level phenotype records	{'is_sample': ['yes'], 'is_random': ['no'], 'source_data': ['Specialty clinics at five North American sites'], 'is_representative': ['no'], 'why_not_representative': ['Cohort-based selection for voice-relevant conditions']}

Subpopulations

Identification
- Adult cohort (v1.0); disease categories include voice disorders, neurological/neurodegenerative disorders, mood/psychiatric disorders, respiratory disorders
Distribution
- Participants selected based on membership in predefined disease cohorts

Distribution Formats

Description
- Parquet
- TSV
- JSON

Distribution Dates

Description
- 2024-11-27

🔍

Collection Process

How was the data acquired?

ID *

bridge2ai-voice-v1.0

Name

Bridge2AI-Voice

Title

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

Description

Bridge2AI-Voice is a multi-site, ethically sourced dataset of human voice linked to clinical and questionnaire information to enable research on voice as a biomarker of health. Version 1.0 includes 12,523 recordings from 306 participants across five North American sites, focusing on cohorts with known voice-relevant conditions (voice disorders, neurological/neurodegenerative disorders, mood/psychiatric disorders, and respiratory disorders). The initial release contains only low-risk, derived data (e.g., spectrograms and acoustic features) and de-identified phenotype tables; original audio waveforms and free-speech transcripts are not included. Data were collected via a standardized protocol using a custom tablet application, with preprocessing that included resampling to 16 kHz with a Butterworth anti-aliasing filter, STFT-based spectrogram generation, extraction of acoustic and phonetic/prosodic features (OpenSMILE, Parselmouth/Praat), and automatic transcription using OpenAI Whisper Large. Documentation: "https://docs.b2ai-voice.org."

Page

https://docs.b2ai-voice.org

Issued

2024-11-27

Keywords

voice
bridge2ai
audio
VOICE

Creators

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

Addressing Gaps

Response
Addresses the lack of large, high-quality, standardized, and demographically diverse voice datasets linked to clinical and questionnaire data across multiple institutions.

Relationships

Description
- Recordings are linked to participants and sessions via participant_id and session_id; tasks identified by task_name

Is Deidentified

Description

HIPAA Safe Harbor identifiers removed (e.g., names, contact details, precise dates, etc.)
State and province removed; country retained
Free speech transcripts removed; original audio waveforms omitted in v1.0

Sensitive Elements

Description
- Contains de-identified health-related information (demographics, clinical variables, questionnaire responses)

Confidential Elements

Description
- De-identified clinical and questionnaire data; identifiable communications removed

Acquisition Methods

Description
- Standardized protocol with voice tasks (e.g., sustained vowel), demographics, health and targeted questionnaires
- Custom tablet application; headset used when possible
- Data exported from REDCap via open-source tooling
Was Directly Observed
yes
Was Reported By Subjects
yes
Was Inferred Derived
yes

Collection Mechanisms

Description
- Custom tablet app and headset for data capture; REDCap used for data entry/export

Data Collectors

Description
- Project investigators at five North American sites; patients presenting at specialty clinics were screened and consented

Ethical Reviews

Description
- Data collection and sharing approved by the University of South Florida IRB; submitted for review to the University of Toronto Research Ethics Board

Preprocessing Strategies

Description
- Raw audio converted to mono and resampled to 16 kHz with a Butterworth anti-aliasing filter
- Spectrograms computed via short-time FFT (25 ms window, 10 ms hop, 512-point FFT)
- Acoustic features extracted with OpenSMILE
- Phonetic and prosodic features computed with Parselmouth and Praat
- Transcriptions generated with OpenAI Whisper Large model
Used Software
Name URL
OpenSMILE
Parselmouth
Praat
torchaudio
OpenAI Whisper Large
b2aiprep https://github.com/sensein/b2aiprep

Name	URL
OpenSMILE
Parselmouth
Praat
torchaudio
OpenAI Whisper Large
b2aiprep	https://github.com/sensein/b2aiprep

Cleaning Strategies

Description
- HIPAA Safe Harbor de-identification (removal of identifiers and fine-grained dates)
- Removal of free speech transcripts
- Omission of original audio waveforms from initial release

Labeling Strategies

Description
- Automatic Transcription Using Openai Whisper Large (note
  free speech transcripts are not included in v1.0)

Raw Sources

Description
- Original audio waveforms collected; not distributed in v1.0. Future releases aim to include voice data with additional precautions.

Maintainers

Description
- Health Data Nexus
- Temerty Centre for AI Research and Education in Medicine

Is Tabular

mixed (Parquet dense arrays and tabular TSV/JSON)

Subsets

Description	Format	ID	Media Type	Name	Path	Title
Parquet dataset containing time-frequency spectrograms derived from voice recordings; includes parti...		spectrograms-parquet	application/x-parquet	spectrograms.parquet	spectrograms.parquet	Spectrograms (derived from voice waveforms)
Tab-delimited file with one row per participant containing demographics, acoustic confounders, and v...		phenotype-tsv	text/tab-separated-values	phenotype.tsv	phenotype.tsv	Phenotype table
JSON data dictionary describing columns in phenotype.tsv; includes a one-sentence description for ea...	JSON	phenotype-json	application/json	phenotype.json	phenotype.json	Phenotype data dictionary
Tab-delimited file with one row per recording containing features derived from OpenSMILE, Praat/Pars...		static-features-tsv	text/tab-separated-values	static_features.tsv	static_features.tsv	Static audio-derived features
JSON data dictionary describing features in static_features.tsv; includes a description for each fea...	JSON	static-features-json	application/json	static_features.json	static_features.json	Static features data dictionary

Motivation

Composition

Collection Process

Uses

Distribution

Maintenance