Global Research and Imaging Platform (GRIP): “FreeSurfer” for Digital Voice Processing
Bibliographic record
Abstract
BACKGROUND: Digital voice is increasingly being recognized as a less biased and more scalable approach for identifying those with cognitive impairment in the early stages of Alzheimer's disease (AD) and other related dementias (ADRD). Just as brain MRI scans were widely adopted as a tool for in vivo detection of AD/ADRD, FreeSurfer was developed as an open-source tool to facilitate brain MRI processing, opening the pathway for analysis and discovery. This toolkit seeks to similarly enable the usage of digital voice for AD/ADRD scientific advancement. METHOD: Global Research and Imaging Platform (GRIP) is developing an open-access platform that will address common research pain points. In collaboration with GRIP, the first version of this toolkit has been published on a public GitHub repository. Table 1 lists the current voice processing tasks that can be accomplished via a variety of open-source packages included in the toolkit, which have been tested on more than 33,000 Framingham Heart Study (FHS) recordings collected between 2005-2024 and on several publicly available datasets. The toolkit contains detailed documentation of examples and containerized deployment via Docker for each package. RESULT: Language identification was performed on the MinDS-14 dataset via existing models (Table 2). Speaker diarization on the VoxConverse dataset via the pyannote.audio package, resulting in the lowest diarization error rate (DER) of 9.0% among several tools. Automatic speech recognition (ASR) pipelines that utilize several available models and test on open-source datasets (MinDS-14, DisfluencySpeech) are included in the toolkit. ASR evaluation pipelines that produce metrics such as word error rate, match error rate, word information lost, word information preserved, and character error rate are also included. Acoustic features such as openSMILE low-level descriptors and audio embeddings (Data2vec, Wav2Vec2) and natural language processing features including pauses and lexical diversity have been produced on FHS recordings (hour-long neuropsychological tests, short smartphone-app based). CONCLUSION: Digital voice may be an ideal scalable option for collection of cognitively relevant information in real-world settings. The ongoing development of this modular toolkit will enable efficient and non-proprietary processing of digital voice. GRIP will allow users worldwide, regardless of their technical expertise, to leverage robust digital voice processing pipelines.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.004 | 0.012 |
| Meta-epidemiology (narrow) | 0.004 | 0.001 |
| Meta-epidemiology (broad) | 0.002 | 0.003 |
| Bibliometrics | 0.003 | 0.002 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.002 | 0.003 |
| Open science | 0.005 | 0.005 |
| Research integrity | 0.002 | 0.003 |
| Insufficient payload (model declined to judge) | 0.046 | 0.054 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".