Voice Biomarkers Explained

    What are voice biomarkers?

    A plain guide to how the human voice carries measurable signals about the body, how those signals are extracted, and where the science still has limits.

    Definition

    Voice, vocal and speech biomarkers

    A voice biomarker is a measurable feature of the human voice that changes in a consistent way with a physiological or psychological state. The terms voice biomarker, vocal biomarker and speech biomarker are often used interchangeably. Researchers sometimes draw a distinction: voice and vocal biomarkers usually describe the sound itself, such as pitch, loudness, breathiness and stability, while speech biomarkers describe how language is produced, such as speaking rate, pauses and articulation.

    In every case the idea is the same. A short recording becomes a set of numbers that can be compared with reference data, tracked over time and studied alongside other health measures. Most voice biomarkers are composite measures: they combine several acoustic and timing features rather than relying on a single number, which makes them more stable from one recording to the next.

    Physiology

    How the voice reflects the body

    Speaking is one of the most complex motor tasks the body performs. It draws on the lungs for airflow, the larynx and vocal folds for sound, the tongue, lips and jaw for articulation, and the brain for planning and timing. Each of these systems can leave a trace in a recording.

    Breathing shapes phrase length and where pauses fall. Muscle tension and hydration affect how the vocal folds vibrate, which shows up as changes in pitch stability, jitter and shimmer. The autonomic nervous system, which responds to stress and fatigue, can influence voice quality and speaking rhythm. Changes in cognitive processing may appear as longer hesitations or reduced fluency.

    These signals are subtle and vary from person to person, so a single feature rarely means much on its own. Patterns across many features, compared with large and varied reference populations, are what make the voice useful as a research measure.

    Method

    How voice biomarkers are extracted

    Extraction usually follows four steps.

    1. 1

      Quality check

      The recording is checked for background noise, clipping, duration and the amount of actual speech.

    2. 2

      Feature measurement

      Signal processing measures acoustic features such as fundamental frequency, intensity, spectral shape and the timing of speech and silence.

    3. 3

      Modelling

      Machine learning models combine these features into composite measures.

    4. 4

      Validation

      Each measure is tested against reference standards collected in controlled studies, and its performance is reported openly.

    At VoiceBiota, this process runs on a dataset of 1,800,637 voice recordings from 54 countries and 35 studies, supporting 46+ validated biomarkers for research and development use. Validation matters as much as extraction: a measure is only useful when its performance is known. Explore the dataset.

    Applications

    Uses in research

    Voice recordings are quick, low cost and non-invasive, and they can be collected remotely over a phone call, an app or a smart speaker. That makes voice attractive to researchers. Current research areas include respiratory health, cardiovascular signals, stress and fatigue, mental wellbeing and cognitive change over time. Because a recording can be repeated as often as needed, voice is also well suited to tracking small changes between clinic visits.

    Common study designs include population studies, longitudinal monitoring where each person acts as their own baseline, and digital endpoints in clinical trials. Developers can also build validated measures into their own products through the VoiceBiota platform.

    Limitations

    What voice biomarkers cannot do

    Voice biomarkers are not diagnostic tools on their own. Results can be affected by recording conditions, microphone quality, language, accent, age, sex, a common cold or simply how someone feels on the day. Models trained on one population may not perform as well on another, which is why diverse reference data and transparent reporting are essential. A good study reports how each measure was validated, who was in the reference group, and how accurate the measure was in people unlike those it was trained on.

    Privacy matters too. A voice recording is personal data, so consent, secure handling and clear limits on storage and reuse are part of responsible research. VoiceBiota measures are for research and evaluation use only and are not intended for clinical diagnosis.

    FAQ

    Frequently asked questions

    Mostly, yes. Both terms describe measurable features of the voice that change with a physiological or psychological state. Speech biomarkers focus more on how language is produced, such as pauses, speaking rate and articulation.

    It depends on the measure. Some acoustic features can be measured from a few seconds of clear voice. Measures of timing, pauses and fluency generally need longer samples of natural speech.

    No. Voice biomarkers are research measures that can show patterns or changes worth investigating. A diagnosis requires a qualified clinician and established clinical tests.

    It can. Some features, such as pitch stability, depend little on language, while speech timing and articulation vary between languages and accents. Reference data from many countries helps account for this.

    A standard microphone, such as the one in a phone or headset, is usually enough. A quiet room and a steady distance from the microphone make results more reliable.

    A voice recording is personal data. Responsible research relies on informed consent, secure processing and clear limits on how long recordings are kept and how they are reused. VoiceBiota processes voice data in line with GDPR.

    ⚠️ Research & Evaluation Use Only: Not for Clinical Use. All voice biomarker values are directional research insights. Not intended for diagnosis or treatment. CE (MDR) certification currently underway. Rezpio is under CE certification for intended use in COPD and Asthma patient management and can currently only be used for research and evaluation purposes.