Smartphone speech test may help assess Huntington’s severity
Automated language analysis predicted clinical scores in small study
Written by |
Smartphone-recorded speech may provide a simple and less burdensome way to assess disease severity in people with Huntington’s disease, a small study suggests.
Researchers found that fully automated analysis of language patterns using speech-recognition and language-analysis tools could statistically predict motor, cognitive, and functional scores with moderate to strong predictive performance. The findings suggest speech-based digital measures could potentially help assess Huntington’s severity remotely, complementing traditional in-clinic assessments.
“Fully automated analysis of smartphone-based language assessment can remotely quantify cognitive, motor, and functional impairment in HD [Huntington’s disease], offering a scalable, low-burden digital biomarker for clinical trials and decentralized monitoring,” the researchers wrote.
The study, “Remote smartphone-based spoken language screening predicts clinical markers in Huntington’s disease,” was published in the Journal of Neural Transmission.
Speech changes may reflect brain changes
Huntington’s is caused by mutations in the HTT gene that lead to the progressive damage and loss of nerve cells in certain parts of the brain. Symptoms, including movement problems, cognitive decline, and behavioral abnormalities, often begin subtly and worsen over time as more nerve cells are lost.
Doctors typically monitor disease progression using the Unified Huntington’s Disease Rating Scale (UHDRS), a standardized tool that assesses motor, cognitive, behavioral, and daily functional abilities.
However, these evaluations require trained specialists and in-person visits, making disease monitoring time-consuming, costly, and potentially less sensitive to subtle or short-term changes.
Because speech relies on brain networks affected early in Huntington’s, researchers have increasingly explored whether changes in how people speak could serve as digital measures of disease severity or progression. But most previous work relied on laboratory testing or manual analysis, limiting their usefulness for frequent monitoring.
Today, automated speech-recognition and language-analysis tools make it possible to study speech using smartphone recordings, without researchers needing to review each recording by hand.
“Yet, whether fully automated, smartphone-based analysis of speech can capture clinically meaningful variation in main HD symptoms remains unknown,” the researchers wrote.
Study tested smartphone speech tasks
To investigate this further, a team of researchers conducted a small, cross-sectional study involving 30 people with Huntington’s (mean age 47.1; 47% men) and 23 healthy volunteers matched by age and sex. The Huntington’s group included 21 Czech and nine German participants: nine were pre-symptomatic, three were prodromal, and 18 had manifest disease. The healthy control group included 17 Czech and six German volunteers.
Huntington’s participants first underwent standard clinical evaluations in the clinic, including assessments of motor symptoms, cognitive performance, functional capacity, and disease stage.
During the initial clinic visit, all participants were provided with a smartphone equipped with a custom application to complete two speech tasks: speaking freely in response to open-ended questions, called a monologue task, and listening to and retelling a short fairy tale in their own words.
After an initial supervised speech assessment in the clinic, participants completed six consecutive days of assessments at home.
Automated speech-recognition and language-analysis tools were then used to transcribe and analyze the recordings for language features previously found to be altered in Huntington’s, including vocabulary range, phrase repetition, sentence length, and grammatical complexity.
Data showed that, overall, people with Huntington’s showed clear differences in how they spoke compared with healthy volunteers.
During the monologue task, Huntington’s participants tended to use a narrower vocabulary, repeat phrases more often, and produce shorter, less grammatically complex sentences. Similar differences in vocabulary range and phrase repetition were also seen during the retelling task.
Language features predicted clinical scores
Notably, speech patterns captured through smartphone recordings statistically predicted clinical scores across several measures of Huntington’s severity, while demographic characteristics such as age, sex, and language alone showed little predictive value.
In both tasks, language features predicted motor impairment, as assessed by the UHDRS Total Motor Score (UHDRS-TMS), and functional capacity, measured using the UHDRS Total Functional Capacity (UHDRS-TFC) scale, with moderate to strong predictive performance.
In the monologue task, speech patterns accounted for 57% of variation in UHDRS-TMS scores across participants and 51% of variation on the UHDRS-TFC scale. Language features also predicted cognitive performance, explaining up to 57% of variation in cognitive test scores.
In the retelling task, language features explained 63% of variation in UHDRS-TMS scores, 59% of variation on the UHDRS-TFC scale, and up to 55% of variation in cognitive test scores.
Models based only on demographic characteristics, such as age, sex, and language, had limited predictive value. Models that also included diagnostic confidence, a measure of how certain doctors were that motor signs were due to Huntington’s, performed better for some measures. Still, adding language features substantially improved predictive performance, with combined models explaining up to 76% of variation in clinical measures when including monologue recordings and up to 81% when retelling-task language features were included.
“This study demonstrates that fully automated linguistic analysis of speech recorded unsupervised at home using a smartphone can capture clinically relevant markers in HD,” the researchers wrote. The findings suggest digital language measures could potentially serve as biomarkers for remotely assessing disease severity. However, the researchers noted that larger studies are needed to validate the results before these tools can be used for long-term disease tracking or treatment studies.
