Longitudinal within-person measurement
The methodological core of the Speech Gap direction, and the reason the planned study is shaped the way it is.
The idea
Absolute speech characteristics differ substantially between individuals. One person’s baseline speaking rate, pause frequency, and pause duration may sit far from another’s without either being clinically meaningful. Comparing patients to each other therefore throws away signal and adds noise.
The informative quantity is instead the participant’s own change over time, relative to their own pre-therapy baseline. Each person becomes their own control.
What follows from this
No untreated control group is needed. A separate untreated control group is not intrinsic to, and is not needed for, this observational question. This is a direct consequence of the within-person design, not a compromise.
Repeated measurement is mandatory, not optional. A single measurement establishes only a reference point. The signal lives in the trajectory, which requires many low-burden measurements — hence weekly short speech samples rather than occasional long sessions.
The instrument must be frozen. If the measurement algorithm changes between week 1 and week 12, an observed change in the numbers cannot be attributed to the participant. This is the single hardest software constraint the clinical design imposes, and it is why PCI Registered Reports and its freezing discipline matter so directly here.
Temporal ordering must be respected. The prediction window must close before the outcome window. A predictor containing information from the future outcome window cannot legitimately be described as an early predictor.
The design principle for data collection
The trend-oriented concept dictates what must be captured alongside the audio: repeatable prompts, timestamps, task and protocol identifiers, recording quality, respiratory-event handling, and links to clinical reference dates. These are what turn short sessions into a longitudinal research asset. → Data governance and annotation
What is claimed, and what is not
Trend detection may establish a causal treatment effect. That is stated as a possibility in the source material, not as an established property of the method.
The use of within-person Speech Gap change to estimate or predict clinical trajectories remains to be tested prospectively against repeated clinical reference assessments. → Editorial safeguards
Related pages
Speech Gap · Temporal speech parameters · PCI Registered Reports · Semmelweis / VIKOTE · Data governance and annotation