Computational framework protects privacy in voice-based cognitive health assessments

Digital speech recordings contain valuable information that may indicate the cognitive health of an individual and offers a non-invasive and efficient method for assessment. Research has shown that digital speech measures can detect early signs of cognitive decline by analyzing characteristics such as speech, articulation, pitch variation and pauses that may indicate cognitive disorders in deviating from normative patterns.

However, speech data introduces privacy challenges because of the personally identifiable information embedded in recordings, such as gender, accent and emotional state, as well as subtle spray brands that can identify individuals unique. These risks are reinforced when speech data is processed by automated systems, so that concern is expressed about Re–Identification and potential abuse of data.

In a new study, Boston University Chobanian & Avedisian School of Medicine researchers have introduced a computational framework that uses pitch-shifting, a sound recording technique that changes the pitch of a sound, increasing or lowering it to protect the identity of the speaker cow.

By using techniques such as pitch-shifting as a way of decrease, we have demonstrated the ability to reduce privacy risks and at the same time retain the diagnostic value of acoustic characteristics. “

Vijaya B. Kolachalama, PhD, Faha, corresponding author, Assistantial teacher Medicine

With the help of data from the Framingham Heart Study (FHS) and Dementiabank Delaware (DBD), the researchers have applied Pitch Shifting at different levels and recorded additional transformations, such as time scale adjustments and addition of noise, to change vocal characteristics on answers on neuropsychological tests. They then assessed the speaker floor via the same error percentage and diagnostic utility through the classification of models for machine learning that distinguish cognitive situations: normal cognition (NC), mild cognitive disorders (MCI) and dementia (de).

With the help of darkened speech files, the computer framework NC, MCI and the differentiation could accurately determine in 62% of the FHS data set and 63% of the DBD data set.

According to the researchers, this work contributes to the ethical and practical integration of speech data in medical analyzes, which emphasizes the importance of protecting the privacy of the patient while maintaining the integrity of cognitive health assessments. “These findings paves the way for the development of standardized, privacy-centric guidelines for future applications of speech-based assessments in clinical and research institutions,” adds Kolachalama, which is also university teacher of computer science, the faculty of the Faculty of Faculty of Facultytitute of Computing Sciscies at Boston University.

These findings appear online in Alzheimer & Dementia: The Journal of the Alzheimer’s Association.

This project was supported by subsidies from the National Institute on Aging’s Artificial Intelligence and Technology Collaboratories (P30-AG073104 and P30-AG073105), the American Heart Association (20SFRN35460031), Gates-Health of Health of Health of Health of Health, Health of Health, Health or the National Institationses of Institutes, R01-AG062109, and R01-AG083735).