Eli Lilly 獲 Alzheimer’s Drug Discovery Foundation 授權使用 SpeechDx 數據集，研究阿茲海默症

The Alzheimer’s Drug Discovery Foundation (ADDF) announced on October 28, 2026 that Eli Lilly and Company has licensed SpeechDx, a longitudinal speech-and-biomarker dataset built by the foundation’s Diagnostics Accelerator, and will use the resource to study whether speech can serve as a tool for Alzheimer’s research. Lilly is the third organization to license the dataset, joining Siemens Healthineers and Callyope.

The foundation characterizes SpeechDx as the largest harmonized longitudinal collection of speech and biomarker data in the world, covering the Alzheimer’s continuum from cognitively unimpaired individuals to Alzheimer’s dementia. In the foundation’s description, digital measures of this kind could be used alongside blood tests to provide additional information on a patient’s cognition and future risk.

Statements From the ADDF, Lilly, and Gates Ventures

ADDF Chief Executive Officer Isobel Coleman called Lilly a world leader in Alzheimer’s therapeutics and diagnostics and described its decision to license the data as recognition of the resource’s quality and of the foundation’s role in advancing speech-based measures. She said AI-driven tools are only as strong as the data behind them, and that SpeechDx is intended to give researchers and industry partners the comprehensive data needed to develop and validate new digital markers.

Mark Mintun, group vice president of Neuroscience Research and Development at Eli Lilly and Company, said the resource combines speech-based measurements with established biomarkers and supports Lilly’s work to make diagnosis more accurate, more accessible, and less invasive at every stage of disease. He said detecting the disease earlier matters because mild symptoms are frequently mistaken for normal aging, which delays opportunities to intervene. Mintun noted that Lilly has worked for more than 35 years to advance the understanding and treatment of Alzheimer’s disease.

Niranjan Bose, Managing Director of Health and Life Sciences at Gates Ventures, said converting speech from a promising signal of cognitive change into a dependable digital measure requires large, well-characterized datasets. He said Lilly’s participation, alongside partners such as Siemens Healthineers and Callyope, moves the effort nearer to establishing speech as a credible, validated instrument for Alzheimer’s research and drug development.

SpeechDx Study Design

The Diagnostics Accelerator launched SpeechDx in 2023, and the announcement describes it as the ADDF’s largest investment in AI to date. The program was created to fill what the foundation describes as a critical gap: the absence of datasets tracking how an individual’s speech changes over time alongside biomarker-confirmed diagnoses and clinical outcomes. The announcement states that the study follows more than 2,000 participants over three years and produces what the foundation calls a gold-standard resource on which AI models can be trained and validated. It also presents speech as among the earliest signs of cognitive decline, noting that ordinary smartphone and computer microphones can capture it at low cost, which the foundation says makes it a highly scalable way to monitor brain health.

The ADDF’s SpeechDx program page describes the study as a multi-site, observational effort following 2,650 participants across global clinical sites. The cohort spans cognitively normal participants and people with subjective cognitive decline, mild cognitive impairment, and Alzheimer’s disease, as well as other conditions including frontotemporal degeneration and Parkinson’s disease. Listed sites include Boston University, Emory University, the Barcelona Brain Health Initiative, the Barcelona Beta Brain Research Centre, the Bogalusa Heart Study, and Ace Alzheimer Center, with data collected in English, Spanish, and Catalan.

Participants contribute speech data quarterly over three years through the SpeechDx app, which administers nine speech-eliciting tasks, including picture description, open-ended questions, story recall, and storytelling, plus baseline assessments covering the PHQ-8 questionnaire, the Karolinska sleepiness scale, and a vigilance assessment. Each participant undergoes neuropsychological testing and MRI imaging and has a defined blood-biomarker amyloid status; many participants additionally undergo PET imaging and have defined cerebrospinal fluid status.

Prior Licensees and Program Background

On September 30, 2025, the Diagnostics Accelerator announced the release of SpeechDx’s first tranche of data and named Callyope, a Paris-based AI healthtech company, as the dataset’s second licensee. Callyope said it would use the data to strengthen an AI speech-foundation model that, in the company’s description, can detect mood, cognitive, psychotic, and motor symptoms from just 30 seconds of speech across mental health and neurodegenerative disorders, including Alzheimer’s and Parkinson’s. The ADDF’s announcements index dates the program’s first licensing announcement to May 1, 2025, when Siemens Healthineers became the first partner to non-exclusively license the dataset.

The Diagnostics Accelerator was launched in 2018 with initial funders ADDF Co-Founder Leonard A. Lauder and Bill Gates to develop novel biomarkers for Alzheimer’s disease and related dementias. On March 17, 2026, the ADDF announced a third phase of the initiative backed by an additional $50 million, which brought the program’s total to $150 million. That phase includes renewed commitments from the initial funders, participation from Biogen, the Dolby family, Eli Lilly and Company, and the Shanahan Family Foundation, and new support from the JTMF Foundation and the Robertson Foundation. The foundation states that the accelerator has invested $100 million across 76 programs worldwide since its founding and that accelerator-funded Fujirebio received the first FDA approval for an Alzheimer’s blood test in 2025.

Data Governance and Access Terms

According to the program page, participants record their quarterly sessions on study-provided tablets, and the audio is encrypted. Because recordings may inadvertently contain identifying details, the encrypted audio is sent to a secure backend where personal identifying information is manually removed, yielding pseudonymized voice data that is transferred to the AD Curation Studio hosted by the Alzheimer’s Disease Data Initiative. Clinical sites contribute their own pseudonymized clinical data to the same environment, where it is paired with the corresponding voice data and harmonized across sites.

A committee reviews and approves every proposed use of the data for compliance with ethical and legal requirements. Approved researchers work with the full harmonized dataset inside a controlled environment within the AD Curation Studio; they cannot download or export the data, and the workspace does not permit audio reproduction software. Because the data was collected for speech-based analyses, the clinical data may not be used for non-speech analysis.

Researchers and companies apply for access, which is non-exclusive, and in exchange the ADDF negotiates investment terms consistent with its venture philanthropy model. The first data became available to SpeechDx partners in Q1 2025, and releases continue in quarterly batches as longitudinal collection proceeds, with the study anticipated to end in Q4 2027.