What is a speech biomarker?
A speech biomarker is a measurable characteristic of how a person speaks, or of what they say, that has been shown to indicate a health state, such as a disease, its severity or its change over time, for a defined context of use.
Explainer for trial sponsors, clinical operations and digital endpoint teams, by Cephalgo · Updated November 2026
From biomarker to speech biomarker
A speech biomarker inherits its meaning from the general definition of a biomarker. What changes is where the measure comes from: a recording of a person speaking.
“A defined characteristic that is measured as an indicator of normal biological processes, pathogenic processes, or responses to an exposure or intervention, including therapeutic interventions.”
Digital biomarker
A biomarker derived from data collected with a digital device, such as a phone, wearable or computer. The VOCAL consensus adds that only digital measures shown to relate to a biological, physiological or clinical state within a defined context of use qualify as digital biomarkers.
Speech biomarker
A digital biomarker taken from recorded speech. It can describe how the sound is made, what is said, or both. VOCAL uses “vocal biomarker” as the overarching term for voice, speech and language, with speech and articulation, and cognition and language, as levels within it.
Context of use
The population, task and purpose a biomarker was shown to work for, such as pre‑screening for an Alzheimer’s trial or tracking speech in Parkinson’s. Evidence in one context does not carry over to another on its own. Under the BEST definitions, a biomarker is not an assessment of how a person feels, functions or survives, so, depending on its use, a regulator may treat a speech task measure as a clinical outcome assessment, such as a performance outcome, as well as or instead of a biomarker.
What a speech biomarker measures
Speech draws on breathing, the voice, the muscles of articulation and the brain’s language and memory systems. Diseases that affect any of them can change speech, so measures fall into a few families.
Acoustic features
How the sound is made: pitch and how much it varies, voice steadiness, breathiness, loudness, and how precisely consonants and vowels are formed. In Parkinson’s, for example, speech can become quieter and flatter in pitch.
Timing features
When sound and silence happen: speech rate, articulation rate, pause length and how often pauses occur, and the rhythm of repeated syllables such as “pa‑ta‑ka”.
Linguistic features
What is said: vocabulary, word choice, sentence length and structure, and the share of vague words such as “thing” or “it”.
Cognitive‑linguistic features
How language serves thinking: word‑finding pauses, how many relevant details of a picture are named, what is recalled from a story, and how many words of a category come to mind in a minute.
The task shapes the measure
Sustained vowels, syllable repetition, reading a passage and free speech are standard motor‑speech tasks (Rusz et al., 2021). Picture description, story recall, verbal fluency and conversation carry more of the language and memory content studied in Alzheimer’s disease (de la Fuente Garcia et al., 2020). Recorded rating interviews hold speech too.
How speech biomarkers are used in CNS clinical trials
Speech can be recorded at a site visit, at home on a participant’s own phone, or found in recordings a trial already holds. That gives four common uses.
Pre‑screening
A short recording before a full screening visit helps decide who is likely to meet the entry range, so fewer candidates are screened and then excluded.
Readings between visits
Short home sessions add readings between clinic visits, so change is read between scheduled visits.
Readings beside a clinical scale
A reading on, or beside, a scale the trial already uses, such as MMSE, MDS‑UPDRS item 2.1 or the ALSFRS‑R speech item, can be joined to each visit score.
Rating interview review
Recorded rating interviews show how each scale was given and scored, which helps central reviewers find rater drift and off‑protocol administration. Rating interview QC ›
Reviews describe uses across diagnosis, risk prediction and remote monitoring (Fagherazzi et al., 2021). In a drug trial, a speech measure tracks a functional scale or symptom; it does not by itself show what a drug does to a biological mechanism.
What makes a speech biomarker ready for a trial
A speech feature becomes useful in a trial when there is evidence for each step from the microphone to the clinical claim, and when the analysis cannot drift during the study.
Verification
Does the capture chain record what it should? In the DiMe V3 framework, verification checks that sensors capture data accurately and consistently. For speech, that means recording quality, the effect of microphones and rooms, and checks for background noise (Robin et al., 2020).
Analytical validation
Does the analysis compute the features accurately and reproducibly? This covers the algorithms that turn audio into measures, test‑retest reliability, and differences by age, sex, accent and education.
Clinical validation
Does the reading correspond to the clinical state in the defined context of use? V3 describes this as showing that a measure identifies, measures or predicts the state of interest in the target population, such as diagnosis, severity or progression.
A fixed analysis
The model, features and version are fixed in the analysis plan before outcome data are read, so the first and last readings of a study come from the same analysis.
Phone and software changes
Operating system updates, noise cancellation and new phones change the audio before analysis. The FDA’s December 2023 guidance on digital health technologies asks sponsors to record the timing and nature of updates, to check that measurements are unaffected, and, where measurements change, to compare data before and after the update. Phone changes and speech endpoints ›
Languages
Speech differs by language and dialect. Multinational trials need the same tasks and quality rules in each language, and calibration on data from that language.
What every speech biomarker study has to plan for
These are properties of speech as a measure. A protocol handles each one by design.
Day‑to‑day variation
Speech also changes with fatigue, mood, medication timing, a cold or a noisy room, so tasks, time of day and quality checks are set in the protocol.
Practice effects
Repeating the same task can improve scores through practice alone, so the analysis plan accounts for practice effects.
Context of use
A reading holds for the context it was validated in: the indication, population, task and language.
Health data
Recorded speech used to measure health is health data, a special category under GDPR, and consent sets what may be analysed later.
Who provides speech biomarkers?
Speech biomarkers for trials come from specialist companies, and speech measures for research also come from academic groups.
Cephalgo keeps a list of the companies whose own sites offer speech biomarkers for clinical trials or studies: Speech biomarker companies for CNS clinical trials ›
Before choosing one, a sponsor can put the same questions to each: How to choose a speech biomarker provider ›
Definitions, guidance and reviews
The definitions and requirements on this page come from these documents.
FDA‑NIH Biomarker Working Group. BEST (Biomarkers, EndpointS, and other Tools) Resource. Glossary.
The regulatory definition of a biomarker used by the FDA and NIH.
U.S. Food and Drug Administration. Digital Health Technologies for Remote Data Acquisition in Clinical Investigations. Final guidance, December 2023.
Verification, validation and fitness for purpose of digital health technologies, and how sponsors handle updates during a trial.
Goldsack JC, et al. Verification, analytical validation, and clinical validation (V3): the foundation of determining fit‑for‑purpose for Biometric Monitoring Technologies (BioMeTs). npj Digital Medicine. 2020;3:55. doi:10.1038/s41746-020-0260-4
The DiMe V3 framework: three stages of evidence a digital measure has to clear.
Pizzimenti M, et al. Consensus‑Based Definitions for Vocal Biomarkers: The International VOCAL Initiative. Digital Biomarkers. 2026;10(1):178–190. doi:10.1159/000553327
International consensus definitions, from biomarker and digital biomarker to voice, speech and language levels.
Robin J, et al. Evaluation of Speech‑Based Digital Biomarkers: Review and Recommendations. Digital Biomarkers. 2020;4(3):99–108. doi:10.1159/000510820
Applies the V3 framework to speech, with recommended checks at each stage.
Fagherazzi G, et al. Voice for Health: The Use of Vocal Biomarkers from Research to Clinical Practice. Digital Biomarkers. 2021;5(1):78–88. doi:10.1159/000515346
Review of vocal biomarker uses in diagnosis, risk prediction and remote monitoring.
Rusz J, et al. Guidelines for Speech Recording and Acoustic Analyses in Dysarthrias of Movement Disorders. Movement Disorders. 2021;36(4):803–814. doi:10.1002/mds.28465
Recording environment, speech tasks and acoustic features for motor speech in movement disorders.
de la Fuente Garcia S, et al. Artificial Intelligence, Speech, and Language Processing Approaches to Monitoring Alzheimer’s Disease: A Systematic Review. Journal of Alzheimer’s Disease. 2020;78(4):1547–1574. doi:10.3233/JAD-200888
Systematic review of speech and language measures for cognitive decline in Alzheimer’s disease.
Speech biomarkers: common questions
What is a speech biomarker?
A speech biomarker is a measurable characteristic of how a person speaks, or of what they say, that has been shown to indicate a health state, such as a disease, its severity or its change over time, for a defined context of use. Speech biomarkers are a kind of digital biomarker: the measure comes from an audio recording of a person speaking.
What is the difference between a speech feature and a speech biomarker?
A speech feature is any value computed from a recording, such as pause length or speech rate. The VOCAL consensus, published in Digital Biomarkers in 2026, notes that only digital measures shown to relate to a biological, physiological or clinical state within a defined context of use qualify as digital biomarkers.
Is a speech biomarker the same as a voice or vocal biomarker?
The terms overlap and are often used interchangeably. The VOCAL consensus uses vocal biomarker as the overarching term for biomarkers from voice, speech and language, and places speech and articulation biomarkers and cognitive and language biomarkers as levels within it.
What does a speech biomarker measure?
Acoustic features describe how the sound is made, such as pitch variation, voice steadiness, articulation and timing. Linguistic and cognitive-linguistic features describe what is said, such as word choice, sentence structure, word-finding pauses and how much of a picture is described. Many measures combine both.
How are speech biomarkers used in clinical trials?
In CNS trials, speech biomarkers are used to pre-screen candidates before a full screening visit, to add readings between clinic visits, to give a reading beside the clinical scale the trial already uses, and to review recorded rating interviews for how each scale was given and scored.
What makes a speech biomarker ready for a clinical trial?
Evidence at three stages of the DiMe V3 framework (verification, analytical validation and clinical validation) for the trial's context of use, an analysis fixed before outcome data are read, a plan for phone and software changes, and calibration for each study language.
Do phone and software changes affect speech biomarkers?
They can. Operating system updates, noise cancellation and new phones change the audio before analysis. The FDA's December 2023 guidance on digital health technologies asks sponsors to record the timing and nature of updates and to check that measurements are unaffected, and, where measurements change, to compare data before and after the update.
Who provides speech biomarkers for clinical trials?
Several companies offer speech biomarkers for clinical trials, alongside academic groups that develop speech measures for research. Cephalgo keeps a list of companies whose own sites offer speech biomarkers for clinical trials or studies.
How Cephalgo works with speech biomarkers
Cephalgo, a speech biomarker company in Strasbourg, France, provides speech biomarkers for Alzheimer’s, Parkinson’s, Huntington’s and ALS trials. Its multi‑stream model reads acoustic‑linguistic features, cognitive‑linguistic features and deep‑learning embeddings from the same recording and returns one reading on, or beside, the scale the study already uses, with the named features behind it. Readings come from recordings a trial already holds, site visits or short home sessions, stay stable across phone and software changes, and are evaluated stage by stage following the DiMe V3 framework. Speech biomarkers ›Validation approach ›