Researchers used a large data set to develop a speech assay capable of approximating expert listener ratings on key parameters of dysarthria in patients with Friedreich ataxia (FA), according to a study published in medRxiv.

Dysarthria is an important characteristic of FA and is closely associated with quality of life. Experts say that speech problems in FA can be condensed into 2 main points: intelligibility on the part of the listener (ie, an inability to understand what is being said) and differing speech (ie, speech that sounds unnatural). 

Current methods for stratifying severity in terms of speech-related problems make use of standardized questions and commands. However, because these tools are subjective, there is significant heterogeneity in terms of how patients perceive the severity of their speech-based pathology. 

Improvements in digital technology, including microphone design and computational processing, have allowed clinical information to be gathered in a more objective manner. The authors of the study thus sought to compare digital speech data with current clinical speech assessments in terms of intelligibility and naturalness of speech. 

Read more about Friedreich ataxia etiology 

The research team acquired data from 132 patients with FA. The span of clinical visits among these participants ranged from a single visit to 4 or more visits over 10 years. Participants completed a number of speech tasks (such as counting numbers and reading a set paragraph). In addition, participants were assessed using the Friedreich Ataxia Rating Scale (FARS) and the speech subscale of the Friedreich Ataxia Impact Scale (FAIS).

The researchers recruited 2 expert listeners blinded to disease features to rate all speech samples for intelligibility and naturalness on a 5-point scale, ranging from unremarkable to severe. To measure the impact of speech on quality of life, the research team used the speech subscale of the FAIS. Speech features were then selected and fed into a random forest and a support vector machine classifier; this was carried out in a standard supervised learning set up developed to replicate scores produced by experts. 

The results of the study demonstrated that a subset of measures strongly correlated with all 4 clinical scales—FARS, FAIS, naturalness, and intelligibility. The mean absolute errors for predicting intelligibility and naturalness were 0.23 and 0.27, respectively, which were lower than the variability observed between subjective listener ratings. 

“Our data suggests that speech working as a multifactorial subset of features can provide an estimate of function within a single decile, and thereby is linked but not wholly driven by disease severity,” the authors concluded. 


Vogel AP, Maruff P, Reece H, et al. Clinically meaningful metrics of speech in neurodegenerative disease: quantification of speech intelligibility and naturalness in ataxia. medRxiv. Published online March 29, 2023. doi:10.1101/2023.03.28.23287878