Acoustic & prosodic
From the sound of speech
- ->Fundamental frequency (F0) mean and variability
- ->Speech rate — words per minute of active speaking
- ->Pause rate and mean pause duration
- ->Jitter, shimmer and harmonics-to-noise ratio
For research
Custom voice and verbal biomarker pipelines.
Bespoke pipelines that extract digital biomarkers from voice and conversation data — acoustic, linguistic and behavioral markers — engineered, versioned and delivered to run on your own data.
Acoustic & prosodicLinguisticBehavioralVersioned & audited
What we extract
Digital biomarkers across three families.
Acoustic, linguistic and behavioral markers, each grounded in the peer-reviewed literature on speech and neurological and affective change.
Linguistic
From the structure and content of speech
- ->Lexical diversity — richness and variety of vocabulary
- ->Syntactic complexity — structural complexity of sentences
- ->Idea density — meaningful information per unit of speech
- ->Discourse coherence — a logical thread across the conversation
Behavioral
From patterns of interaction
- ->Conversational initiative — who starts and drives exchanges
- ->Turn-taking — participation based on turn counts, not word counts
- ->Engagement quality across the conversation
Pipeline stages
From raw audio to research-ready markers.
Every stage is engineered and operated in-house, with explicit versioning at each step and a full audit trail of how each data point was produced.
Stage 01
Ingest
Audio is ingested from your source — archived calls, telemedicine, contact-center recordings or any speaker-separated feed — with lineage recorded from the first step.
Stage 02
Normalize
Disfluencies and turn-taking are normalised and language is detected and propagated downstream, so markers are computed on a consistent basis.
Stage 03
Speaker-separate
Each speaker's audio is isolated, so a companion's or clinician's speech never contaminates the participant's signal.
Stage 04
Extract
Acoustic, linguistic and behavioral markers are computed — some deterministically, some through semantic evaluation — before any scoring.
Stage 05
Score vs baseline
Each marker is compared to the participant's own baseline, producing standardized deviations and composite indices that track change over time.
Stage 06
Export
Structured records are exported in your chosen format, with a versioned data dictionary and full audit lineage attached.
Works on your data
Runs on the audio you already have.
The pipeline does not require a specific capture app or device — it works on any speaker-separated audio source.
Archived call audio
Batch-analyze historical recordings to build retrospective marker trajectories from data you already hold.
Telemedicine
Extract markers from remote visits, augmenting the clinical record with objective voice-derived signal.
Contact-center
Turn recurring support or check-in calls into a longitudinal signal layer without changing the interaction.
Any speaker-separated source
Point the pipeline at any audio channel that produces speaker-separated speech, and it will extract the marker set you specify.
Scientific grounding
Grounded in peer-reviewed research.
Speech-derived biomarkers are an established and rapidly maturing field. The markers we extract are the ones the literature validates.
0.81
AUC for identifying individuals below normative cognitive thresholds from speech (npj Digital Medicine, 2026)
0.873
F1-score discriminating cognitive impairment from healthy controls in conversational speech (CognoSpeak, 2025)
88%
Accuracy distinguishing early Alzheimer's from controls using linguistic features (Frontiers in Aging Neuroscience, 2024)
The same marker families power CogPath and VGDS, and feed the multimodal CNS R&D datasets.
Deliverables
What you receive.
A pipeline you can rely on and defend — versioned, documented and traceable end to end.
Versioned pipeline
The extraction pipeline is delivered with explicit version pinning, so every run is reproducible and every change is traceable.
Data dictionary
A versioned dictionary documents every field, its type, its units and how it is derived.
Audit lineage
Every record carries a full trail of how each data point was produced, from source audio to final marker.
Open export formats
Structured outputs are delivered as JSON, CSV or Parquet, ready to load into your analysis environment.
A pipeline often pairs with data collection — see managed and bespoke Voice AI for decentralized trials.
Discuss a pipeline build.
Tell us about your data source and the markers your program needs, and we will scope a bespoke extraction pipeline — engineered, versioned and delivered to run on your data.