Prediction of Speech Impairment in Patients Treated for Oral or Oropharyngeal Cancer Using Automatic Speech Analysis.
Mathieu Balaguer, Julien Pinquier, Jérôme Farinas and 1 others
PMID 40755165WHAT IT FOUND
Automatic analysis of spontaneous speech predicted patient-reported communication impairment in 25 patients treated for oral or oropharyngeal cancer.
Two phonemic rates and self-reported psychosocial factors explained most variance. The sample was small and the tool is not yet ready for routine clinical use.
Key findings
01Two automatic phonemic parameters, the number of occlusives and sonants recognised per second, predicted the patient-reported Holistic Communication Score with a cross-validation correlation of 0.82.
02Adding eight self-reported biopsychosocial items to the two automatic parameters increased the variance explained in the communication score to 83.6%, but reduced five-fold cross-validation accuracy.
03The automatic speech recognition system performed poorly on this population, with a word error rate of 84.51%, limiting the usefulness of higher-level linguistic parameters.
STILL TO COME
How it was doneWhat they foundWhat it means for SLPs
Read the rest of this summary
You get three full summaries a month, free, and we do not ask for a card. Search, the TL;DRs and your library stay unlimited either way.
What it does not show
The sample size was very small (25 participants), which limits the reliability of the predictive models and the generalisability of the findings. The automatic speech recognition system used was trained on healthy adult speech and performed poorly on this clinical population, with a word error rate of 84.51%. This means higher-level linguistic features were likely misinterpreted. The study is cross-sectional, so it cannot show whether the automatic analysis predicts future communication outcomes or how they change over time. The model combining automatic and self-reported factors had better fit on the training data but worse cross-validation performance than the automatic-only model, raising questions about its practical utility. Participants were recruited from a single ENT department in France, and the speech analysis relied on French-specific tools, which may not transfer to other languages or settings.
Declared interests
The authors declare no conflicts of interest.
The easy way to misread this
Do not interpret the high correlation between automatic parameters and patient-reported impairment as evidence that this tool is ready for clinical use. The automatic speech recognition system had an 84.51% word error rate on this population, and the small sample size means the predictive models are not yet validated for routine practice.
Summarised by AI from the full paper, without a clinician reviewing it. Check it against the source before it changes what you do. Read it on PubMed →