Detecting Cognitive Impairment Using Speech Analysis
Author Information
Author(s): Agbavor Felix, Liang Hualou, Tafiadis Dionysus, Ziavra Nafsika, Voniati Louiza, Papadopoulos Angelos
Primary Institution: School of Biomedical Engineering, Science and Health Systems, Drexel University
Hypothesis
Can multilingual speech analysis effectively predict cognitive impairment in individuals?
Conclusion
The study shows that language-specific ensemble modeling with Whisper embeddings can accurately assess cognitive health in multilingual settings.
Supporting Evidence
- The model achieved an unweighted average recall of 81.83% in MCI classification.
- The root mean squared error for cognitive score prediction was 1.196.
- The ensemble approach outperformed language-agnostic methods in all metrics.
- Majority voting improved prediction accuracy across different tasks.
- Language-specific models captured important nuances for cognitive impairment prediction.
Takeaway
This study found a way to tell if someone has memory problems by listening to how they speak, using special computer programs that understand different languages.
Methodology
The study used speech samples from English and Chinese speakers to classify mild cognitive impairment and predict cognitive scores using a speech foundation model called Whisper.
Potential Biases
The dataset was balanced across age and gender, but potential biases related to education and culture in MMSE assessments remain.
Limitations
The study only tested English and Chinese, limiting generalizability to other languages, and relied on the MMSE, which has known limitations.
Participant Demographics
Participants included 129 individuals with mild cognitive impairment and normal cognition, balanced across age and gender.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website