Automatic prediction of intelligible speaking rate for individuals with ALS from speech acoustic and articulatory samples.
Jun Wang, Prasanna V Kothalkar, Myungjong Kim and 6 others
PMID 30409057WHAT IT FOUND
A model estimated intelligible speaking rate from a single short speech sample.
Combining sound with lip and tongue movement gave the closest estimates, but it was tested in only 12 people with ALS and 2 controls.
Key findings
01The model predicted intelligible speaking rate from a single short speech sample using acoustic and articulatory features.
02Prediction was best when acoustic, lip, and tongue movement information were used together (R 2 = 0.712; RMSE = 37.51 WPM).
03The study used 25 sessions from 12 individuals with ALS and 2 healthy controls, yielding 1,832 valid phrase samples.
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 study included only 12 individuals with ALS and 2 healthy controls, so the model was not tested in a large clinical population. The prediction was evaluated by splitting the same data set into three parts, not by testing it on a separate group of new patients. The reference intelligible speaking rate was based on manual marking by only one speech-language pathologist, which may bias comparison with machine prediction. The paper states that the data set contains a relatively small number of patients and that larger studies are needed to refine and validate the approach. Recording articulatory data with sensors is currently logistically difficult in clinical settings.
Declared interests
The supplied text does not report author conflicts of interest. Publication types indicate research support from the National Institutes of Health and non-U.S. government sources.
The easy way to misread this
Do not treat this as a validated clinical test or as proof that ALS speech decline can be diagnosed from a short sample. The model was built on a small set of sessions and tested by splitting the same data, not by evaluating it in new patients in routine care.