Rubric20-Semantic Evaluation Report

All mandatory fields present with exceptional comprehensiveness. Description is 540+ characters providing detailed project overview. Keywords cover disease categories, technical terms, tools, and platforms. License extensively documented.

id: https://doi.org/10.13026/37yb-1t42, title: Bridge2AI-Voice - An ethically-sourced..., description: 540+ chars comprehensive, keywords: 47 keywords, license_and_use_terms: extensively detailed with 12 specific terms

Fields semantically validated - DOI uses correct PhysioNet prefix (10.13026), title accurately describes dataset characteristics, description provides specific quantitative details (306 participants, 12,523 recordings, five sites)

Narrative fields exceed adequacy threshold significantly. Description provides detailed project context with specific participant counts, disease categories, and technical specifications. Purpose statements comprehensively document objectives.

description: 540+ chars, purposes: 3 entries averaging 220 chars each providing comprehensive motivation documentation

Content semantically rich with specific details rather than generic statements. Descriptions provide measurable outcomes and technical specifications appropriate for biomedical dataset documentation.

Exceptional keyword diversity covering clinical domains, technical methodologies, specific diseases, ethical frameworks, and distribution platforms. Significantly exceeds threshold for excellent discoverability.

keywords: 47 unique terms including disease categories (voice disorders, neurological disorders, mood disorders, respiratory disorders, pediatric), technical terms (spectrogram, MFCC, OpenSMILE, Praat, Parselmouth), specific conditions (Parkinson's, Alzheimer's, depression, COPD), principles (FAIR, CARE), and platforms (PhysioNet, Health Data Nexus)

Keywords semantically appropriate and specific. Disease terms align with documented subpopulations, technical terms match preprocessing strategies, platform names correspond to external resources. No generic or mismatched keywords detected.

Multiple file types supporting different data modalities. Parquet for large numerical arrays (spectrograms, MFCCs), TSV for tabular data (phenotypes, features), raw audio for primary data. Indicates comprehensive multi-modal dataset.

distribution_formats: 5 formats documented - Parquet for spectrograms, Parquet for MFCCs, TSV for phenotype data, TSV for static features, Raw audio (controlled access). Multiple data modalities represented.

Format choices semantically appropriate for data types. Parquet optimal for high-dimensional numerical arrays, TSV with JSON dictionaries standard for tabular clinical data, controlled access for raw audio aligns with biometric sensitivity.

Instance counts comprehensively documented with participant and recording counts. Version-specific metrics provided showing dataset growth over releases.

instances: 306 participants with 12,523 recordings documented. Source file metadata: 89K, 9 source files. Specific counts provided for initial release (v1.0) and subsequent versions.

Instance counts semantically consistent across document. 306 participants and 12,523 recordings cited consistently in description, instances section, and version history. Numbers plausible for multi-year multi-institutional study.

Multiple persistent identifiers provided including primary DOI, project documentation URL, and secondary DOIs for related publications and software releases. Enables robust citation and discovery.

id: https://doi.org/10.13026/37yb-1t42 (PhysioNet DOI), page: https://docs.b2ai-voice.org, external_resources: multiple persistent URLs including PhysioNet landing page, GitHub repository, NIH RePORTER, Zenodo DOI (10.5281/zenodo.13834653), Interspeech publication DOI (10.21437/Interspeech.2024-1926)

DOI 10.13026/37yb-1t42 validated - 10.13026 is correct PhysioNet registrar prefix managed by MIT LCP. Zenodo DOI 10.5281 uses correct Zenodo prefix. Publication DOI 10.21437 uses correct Interspeech conference prefix. All URLs follow valid patterns.

Exceptional funding documentation with complete grant details, budget breakdown, administering institute, opportunity number, and study section. Multiple creators listed with institutional affiliations and roles. Demonstrates comprehensive acknowledgement of contributions.

funders: 2 detailed entries - (1) NIH Office of the Director with grant 3OT2OD032720-01S3 including opportunity number OTA-21-008, project dates, funding amounts (total $4,660,942, direct $4,072,321, indirect $588,621), study section DCMM; (2) NIBIB grant R01EB030362 for PhysioNet. creators: 13 entries with names and institutional affiliations including Contact PI (Yael Bensoussan, USF) and 12 co-investigators with role descriptions.

Grant 3OT2OD032720-01S3 validated - follows NIH supplement format (supplement revision 3 to parent grant OT2OD032720-01). Funding amount $4.66M plausible for large multi-institutional Bridge2AI consortium. Date range Sept 2022 to Nov 2026 consistent with version release timeline. R01EB030362 validated as standard NIH R01 format for NIBIB.

Exemplary ethical documentation covering all major protection areas. IRB approval specified, informed consent process detailed, deidentification methodology comprehensive (HIPAA Safe Harbor with explicit category list), vulnerable population considerations addressed (pediatric cohort with additional privacy precautions), Certificate of Confidentiality provides legal protection against compulsory disclosure.

human_subject_research: extensively documented with USF IRB approval, involves_human_subjects=true, written informed consent described. sensitive_elements: 4 detailed entries covering voice as biometric identifier, EHR data linkage, demographic de-identification, Certificate of Confidentiality. cleaning_strategies: HIPAA Safe Harbor de-identification with all 18 identifier categories enumerated. Privacy protections include raw audio controlled access, free speech transcript removal, geographic data limitation (state/province removed, country retained).

Consistency checks passed - human_subject_research=true aligned with IRB documentation and informed consent details. Deidentification method (HIPAA Safe Harbor) appropriate for health data type and comprehensively documented with all 18 categories. Privacy approach (controlled access for raw audio, derived features only in public dataset) consistent with voice as biometric identifier. Certificate of Confidentiality correctly described with legal protections against court orders and subpoenas.

Exceptional governance documentation with comprehensive license terms, clear access requirements differentiated by data sensitivity level (public registered access for derived features, controlled access for raw audio), specific restrictions on sharing and use, retention and disposition requirements, and legal protections via Certificate of Confidentiality.

license_and_use_terms: extensively documented with name (Bridge2AI Voice Registered Access License), detailed description of access mechanisms (PhysioNet registered access vs DACO controlled access for raw audio), 12 specific license terms including registered access requirement, DUA signature, authorized personnel restrictions, no third-party sharing, safeguards requirements, Certificate of Confidentiality protections, two-year term. Multiple access restrictions documented: registered access required, Data Use Agreement mandatory, controlled access for raw audio via DACO application.

License terms semantically consistent with access mechanisms - registered access for derived features aligns with HIPAA Safe Harbor de-identification, controlled access for raw audio aligns with voice as biometric identifier. Two-year retention term matches typical DUA periods. Confidentiality level implicitly high based on Certificate of Confidentiality coverage and controlled access requirements.

Strong interoperability through use of standard file formats (Parquet for numerical arrays, TSV for tabular data, JSON for metadata), standard acoustic processing pipelines (STFT, MFCC, OpenSMILE, Praat), and schema documentation via JSON data dictionaries. Compatible with Python datasets library and common data science tools explicitly noted.

distribution_formats: Standard formats documented - Parquet (Apache Parquet columnar format for large arrays), TSV (tab-delimited text, standard tabular format), JSON data dictionaries accompanying TSV files. preprocessing_strategies: Standard audio processing (16 kHz sampling rate, monaural, Butterworth filter), standard acoustic features (spectrograms via STFT, MFCCs, OpenSMILE feature set, Praat/Parselmouth phonetic features). Schema conformance: Data exported from REDCap using b2aiprep library, phenotype.json and static_features.json data dictionaries provide schema documentation.

Format choices semantically appropriate and standards-aligned. Parquet is Apache standard for columnar data storage. TSV with JSON dictionaries follows common practice for tabular datasets with metadata. Audio processing parameters (16 kHz sampling, FFT parameters) align with speech processing standards. OpenSMILE and Praat are established acoustic analysis tools with standardized feature sets.

Exceptional technical documentation with comprehensive preprocessing pipeline documentation including specific parameters (window sizes, FFT points, MFCC counts, sampling rates). Software tools clearly named. b2aiprep library has GitHub link in external_resources. Whisper model specified as 'Large' variant.

preprocessing_strategies: 7 detailed entries documenting audio resampling (Butterworth filter), spectrogram extraction (STFT with 25ms window, 10ms hop, 512-point FFT), MFCC extraction (60 coefficients), OpenSMILE features, Parselmouth/Praat phonetic features, Whisper Large transcription, b2aiprep REDCap export. cleaning_strategies: 3 entries covering HIPAA Safe Harbor de-identification, privacy protection measures, multi-site standardization. Software tools mentioned: OpenSMILE, Praat, Parselmouth, OpenAI Whisper Large model, b2aiprep library (with GitHub link in external_resources), REDCap, Python datasets library.

Software tools semantically appropriate for voice biomarker research. OpenSMILE is established speech analysis toolkit. Praat/Parselmouth are standard for phonetic analysis. Whisper Large is state-of-the-art ASR model. b2aiprep is project-specific open source library with GitHub link provided. Processing parameters (16 kHz sampling, 512-point FFT, 60 MFCCs) align with speech processing best practices.

Comprehensive collection protocol documentation with specific mechanisms (custom smartphone app, headset use), acquisition methods covering multiple data modalities (voice tasks, questionnaires, EHR linkage), multi-institutional sites (five specialty clinics), standardized protocols, and detailed timeline with version-specific release dates.

collection_mechanisms: 3 detailed entries describing custom smartphone application with headset, standardized protocol across sites, single vs multiple session requirements, multi-institutional collection across five specialty clinics, screening and consent process, enrollment 2022-2026. acquisition_methods: 3 entries covering voice recording tasks (sustained phonation, respiratory sounds, cough, free speech), self-report questionnaires (demographics, medical history, disease-specific validated instruments, acoustic confounders), EHR access for gold standard validation. collection_timeframes: enrollment 2022-2026, version releases documented (v1.0 Jan 2025, v1.1 Jan 2025, v2.0.0 Apr 2025, v2.0.1 Aug 2025).

Collection protocol semantically consistent and comprehensive. Multi-institutional enrollment across five sites aligns with funders description of five disease cohorts and sampling_strategies. Timeline (2022-2026) matches grant period (Sept 2022 to Nov 2026). Version releases (Jan-Aug 2025) plausible for data collected starting 2022. Custom smartphone app with standardized protocol appropriate for multi-site consistency.

Exemplary version history with specific release dates, version numbers following semantic versioning, descriptions of what changed in each version (features added, cohort expanded), ongoing update plans with end date, version-specific vs latest-version DOI access documented. Demonstrates mature data management practices.

updates: Comprehensive documentation of versioned releases - v1.0 (Jan 17, 2025: 306 participants, 12,523 recordings), v1.1 (Jan 17, 2025: added MFCC features), v2.0.0 (Apr 16, 2025: expanded cohort), v2.0.1 (Aug 18, 2025: latest version). Version access: DOI for latest version (https://doi.org/10.13026/37yb-1t42) with version-specific DOIs mentioned. Update plans: ongoing data collection through Nov 30, 2026, future releases planned with additional participants and pediatric cohort. Frequency: periodic versioned releases during data collection period.

Version timeline semantically consistent with collection timeframe (2022-2026) and grant period. Release dates (Jan-Aug 2025) plausible for data collection starting 2022. Version numbering appropriate (1.0 initial, 1.1 feature addition, 2.0.0 major expansion, 2.0.1 minor update). Future plans (pediatric cohort, raw audio access with additional security) align with documented subpopulations and privacy considerations.

Extensive external resource documentation with multiple DOI-linked publications (Zenodo, Interspeech), code repositories (GitHub for b2aiprep and documentation), dataset distribution platforms (PhysioNet, Health Data Nexus), funding information (NIH RePORTER), and project documentation. Demonstrates comprehensive scholarly and technical integration.

external_resources: 11 entries including PhysioNet landing page, project documentation (docs.b2ai-voice.org), GitHub repository (bridge2ai-docs), NIH RePORTER project details, Health Data Nexus, Zenodo archive (10.5281/zenodo.13834653), Interspeech 2024 protocol publication (10.21437/Interspeech.2024-1926), DACO contact, Bridge2AI program site, b2aiprep software library GitHub. Multiple DOI-linked references provided.

External resources semantically appropriate and cross-validated. Interspeech 2024 DOI (10.21437) uses correct conference prefix. Zenodo DOI (10.5281) uses correct Zenodo prefix. GitHub repositories link to specific projects (eipm/bridge2ai-docs, sensein/b2aiprep). NIH RePORTER project ID (11376382) corresponds to documented grant. PhysioNet URL matches DOI landing page. Multi-platform presence (PhysioNet, Health Data Nexus, Zenodo) demonstrates FAIR distribution practices.

Exceptional human subject characterization with specific participant counts (306 participants, 12,523 recordings), detailed subpopulation definitions across five disease categories with specific conditions listed (e.g., Parkinson's, Alzheimer's, depression, COPD, autism), recruitment strategy clearly articulated (specialty clinic screening, targeted enrollment), and demographic considerations (adult cohort currently, pediatric planned with additional ethics precautions).

instances: 306 participants with detailed description of recruitment from five specialty clinics across five disease cohorts (Respiratory, Voice, Neurological, Mood, Pediatric), multi-institutional sites in North America, standardized protocols, adult cohort in v1.1. subpopulations: 5 detailed entries characterizing Voice Disorders cohort, Neurological/Neurodegenerative cohort, Mood/Psychiatric cohort, Respiratory cohort, Pediatric cohort with specific disease examples for each. sampling_strategies: targeted recruitment from specialty clinics, screening based on conditions manifesting in voice, membership in five predetermined disease cohorts.

Subpopulation characterization semantically consistent with purposes, tasks, and keywords. Five disease cohorts (Voice, Neurological, Mood, Respiratory, Pediatric) align with keywords (voice disorders, Parkinson's, depression, COPD, autism) and tasks (predictive models for these conditions). Specialty clinic recruitment appropriate for disease-specific cohorts. Participant count (306) plausible for multi-year multi-site study. Note that pediatric data not yet released in v1.1 consistent with additional ethics considerations mentioned.

Excellent findability with primary DOI, project documentation URL, and extensive cross-platform linking to PhysioNet, Health Data Nexus, Zenodo, GitHub, and NIH databases. Persistent identifiers enable robust citation and discovery across multiple platforms.

page: https://docs.b2ai-voice.org, id: https://doi.org/10.13026/37yb-1t42, external_resources: 11 persistent URLs including PhysioNet landing page (https://physionet.org/content/b2ai-voice/), GitHub repositories, Health Data Nexus (https://healthdatanexus.ai/content/b2ai-voice/1.0/), Zenodo (https://doi.org/10.5281/zenodo.13834653), NIH RePORTER (https://reporter.nih.gov/project-details/11376382), Interspeech publication DOI, Bridge2AI program site, b2aiprep library GitHub.

All URLs follow valid patterns. DOI resolves to PhysioNet landing page. Multi-platform presence (PhysioNet, Health Data Nexus, Zenodo) demonstrates FAIR distribution. GitHub links provide access to open source tools (b2aiprep) and documentation. NIH RePORTER link enables grant information discovery.

Exceptional access mechanism documentation with clear differentiation between public registered access (PhysioNet with DUA) and controlled access (DACO application for raw audio), specific URLs and contact information provided, platform infrastructure detailed (PhysioNet managed by MIT LCP under NIBIB grant R01EB030362), and step-by-step requirements outlined (DUA signature, authorized personnel, safeguards).

distribution_formats: 5 detailed entries each with access_urls field specifying PhysioNet landing page (https://physionet.org/content/b2ai-voice/) for public datasets or contact email (DACO@b2ai-voice.org) for controlled access raw audio. license_and_use_terms: Comprehensive description of access mechanisms - PhysioNet registered access requires DUA signature, authorized personnel listing, controlled access for raw audio requires DACO application with formal vetting. Platform explicitly named (PhysioNet managed by MIT LCP).

Access mechanisms semantically consistent with data sensitivity levels. Derived features (spectrograms, MFCCs, phenotypes) available via PhysioNet registered access aligns with HIPAA Safe Harbor de-identification. Raw audio requiring controlled DACO access aligns with voice as biometric identifier requiring additional privacy protection. Two-tier access model (public registered vs controlled) appropriate for biomedical dataset with varying sensitivity levels.

Highly detailed license with explicit reuse terms covering permitted uses (research, publication), required practices (DUA signature, safeguards, source recognition), prohibited actions (unauthorized sharing, third-party distribution), and specific conditions (two-year term, authorized personnel only). Encourages open-access publication of results while protecting participant privacy.

license_and_use_terms: 12 specific license terms documented - registered access required, DUA signature mandatory, use restricted to authorized persons, no third-party sharing without consent, safeguards required, compliance with laws and professional standards, public disclosure of results encouraged in open-access journals, recognition of data source required in publications, Certificate of Confidentiality protections apply, two-year term. Permitted uses: research, publication in open-access journals. Prohibited uses: sharing without consent, unauthorized personnel access.

License terms semantically consistent with research dataset purpose. Encouragement of open-access publication aligns with FAIR principles and public good mission. Restrictions (no sharing, authorized personnel only, safeguards required) appropriate for health data with Certificate of Confidentiality. Two-year term standard for research DUAs. Recognition requirement supports proper citation and reproducibility.

Exceptional provenance tracking with structured version control including semantic version numbers, specific release dates, change descriptions for each version, version-specific access mechanisms, and comprehensive generation metadata documenting method, source files, schema, and timestamp. Demonstrates mature data integrity practices.

updates: Comprehensive version history with specific release dates - v1.0 (Jan 17, 2025), v1.1 (Jan 17, 2025), v2.0.0 (Apr 16, 2025), v2.0.1 (Aug 18, 2025). Change log documented with descriptions of what changed in each version (initial release 306 participants/12,523 recordings, added MFCC features, expanded cohort, latest version). Version access documented with DOI for latest version and version-specific DOIs available. Generation metadata included at file header: generation method (Claude Code Agent Deterministic), source file (VOICE_preprocessed.txt with size 89K, 9 source files), schema version, generation date (2025-12-20).

Version control semantically robust with appropriate versioning scheme (1.0 initial, 1.1 feature addition, 2.0.0 major expansion, 2.0.1 minor update). Release dates chronologically ordered and plausible. Change descriptions specific and measurable (participant/recording counts, feature additions). Generation metadata provides full reproducibility chain from source files through processing to final D4D file.

Excellent cross-platform interlinking with 11 external resources spanning data repositories (PhysioNet, Health Data Nexus, Zenodo), code repositories (GitHub), publication databases (Interspeech DOI), funding databases (NIH RePORTER), and program sites (Bridge2AI). Demonstrates comprehensive integration into research ecosystem.

external_resources: Cross-platform links to PhysioNet (primary distribution), Health Data Nexus (alternative repository with version-specific URL 1.0), Zenodo (software/documentation archive with DOI 10.5281/zenodo.13834653), GitHub (two repositories: eipm/bridge2ai-docs for documentation, sensein/b2aiprep for preprocessing library), NIH RePORTER (grant database with project ID 11376382), PhysioNet platform (research resource for physiologic signals), Bridge2AI program site (parent initiative), Interspeech 2024 (protocol publication with DOI), DACO contact (controlled access portal).

Cross-platform links semantically validated and appropriate. PhysioNet is primary distribution platform for biomedical signals data. Health Data Nexus provides alternative access with version-specific URL (1.0) consistent with version history. Zenodo for software/documentation archives aligns with open science practices. GitHub repositories provide open source tools (b2aiprep) and documentation dashboard. NIH RePORTER link connects to funding information. Multiple platforms enhance FAIR compliance and discoverability.