ISCTM 2026 · Philadelphia · 16 September 2026 · Poster
A phone update can move a speech endpoint.
Test for it when you validate.
Cephalgo’s ISCTM 2026 poster, Philadelphia, September: phone software updates can change the audio and shift speech measures in patients who have not changed, a risk to test for when validating a speech endpoint.
When the device changes mid‑trial: speech‑based digital endpoint robustness to on‑device preprocessing variability
Accepted abstract: “Preprocessing variability as an analytical‑validation dimension for speech‑based Parkinson’s disease endpoints”.
- Conference
- ISCTM 2026 Autumn Conference, International Society for CNS Clinical Trials and Methodology
- Dates
- 16–17 September 2026
- Venue
- The Westin Philadelphia, Philadelphia, PA, USA
- Session
- Formal poster session, Wednesday 16 September 2026, Grand Ballroom Salon III
- Data
- A public Parkinson’s speech corpus, Parkinson’s disease versus healthy controls
The speaker stays the same. The audio does not.
The same Parkinson’s and control recordings were run through different noise‑cancellation methods, standing in for successive phone updates. Only the processing step differed.
Software updates reach the audio
Over months or years, participants’ phones receive software updates that can change built‑in noise cancellation, so the audio changes before any analysis sees it.
Change with no clinical cause
Speech measures can then move for a person whose clinical state has not changed, and nothing in the data flags it.
One reading from three streams
Cephalgo’s multi‑stream model reads each recording three ways: language, voice acoustics and a learned audio representation. Its reading stayed stable across recording conditions and, even in its weakest condition, distinguished patients from controls better than any other approach tested.
Published panels can move too
A reconstructed voice‑feature panel of the kind proposed as an endpoint also moved under the same processing change. The poster tested one published combination; other panels can be checked the same way.
Add stability to how you validate
The accepted abstract frames stability under recording‑processing changes as part of validating any speech endpoint. The check can run on recordings you already hold, beside an existing single‑measure endpoint.
A stability audit
Where phone or software changes can enter your study, and whether your current speech measures move with them. Stability audit ›
Your own or your vendor’s speech measures
See whether the measures you already use hold up when the audio processing changes. Already using a vendor ›
How each reading is tested
Reproducibility when phones or their software change is one stage of how Cephalgo evaluates every reading. Validation approach ›
Parkinson’s programs
What Cephalgo reads in de novo and early Parkinson’s trials. Parkinson’s ›
From Copenhagen to Boston
Five posters in 2026, from explainable screening in Copenhagen to phone and software changes in Basel, Philadelphia and Boston.
- March 2026
AD/PD™ 2026, Copenhagen · two ePosters
Validation of fully explainable speech biomarkers for AD/MCI screening, and for PD/HC screening
More about these ePosters › - June 2026
Digital Biomarkers in Clinical Trials Summit, Basel
When the device changes mid‑trial: evaluating speech‑based digital endpoint robustness to on‑device preprocessing variability in longitudinal remote monitoring
More about this poster › - September 2026
ISCTM Autumn Conference, Philadelphia
When the device changes mid‑trial: speech‑based digital endpoint robustness to on‑device preprocessing variability
You are here - November 2026
CTAD 2026, Boston · P446
Noise‑cancellation variability as a source of longitudinal drift in speech‑based digital endpoints for Alzheimer’s disease trials
More about this poster ›
