Seven to ten million people worldwide are suffering from the debilitating neurological disease Parkinson, says Parkinson Disease foundation.
It is a progressive disease that gets worse over time and kills the nerve cells that generate dopamine – a chemical that sends signal to part of our brain that controls movement.
And by the time a person recognizes symptoms of Parkinson, many of his dopamine producing nerve cells have already died. And on top of that, there is no blood test that can detect it and at present, the disease has no cure.
In 2012, Dr. Max Little, an applied mathematician, disclosed his machine learning algorithm that can detect signs of Parkinson by analyzing a 30 second voice recording. And his algorithm is 99 percent accurate.
And now Samsung wants to keep the ball rolling by introducing a voice health diagnosing feature to detect Parkinson from voice of a smartphone user, hints one of their recent patents.
Not only this, Samsung is using vocal analysis to also detect laryngeal cancer and occipital pain – a headache disorder whose causes are unknown.
The whole system consists of four parts: a voice detector, a voice analyzer, a voice diagnoser and a health reporter.
The voice detector, which contains an analog-to-digital converter, monitors the changes in a smartphone user voice. It monitors and detects a user’s voice through a phone conversation and at times when he records his voice.
The voice analyzer selects a type of health state to be diagnosed; for example laryngeal cancer or Parkinson’s disease. After that, on the basis of the type of health state, the analyzer selects the voice period for detecting any abnormalities.
For example, in case of laryngeal cancer, only vowel portions are selected, and in the Parkinson’s disease, an entire sentence may be selected.
In the next step, the analyzer compares a database of a normal voice and an abnormal voice to find out any abnormality in a user’s voice. If abnormality detected, it reports it to the voice diagnoser.
The voice diagnoser, in turns, monitors the change in the user’s voice over time. In the next step, it determines the state of a user health and sends the report to the health reporter that further outputs a warning on the basis of the results.
The patent mentions using various voice features like Jitter (JITT), Shimmer (SHIMM), and Harmonics-to-Noise Ratio (HNR) to detect abnormality in a user’s voice.
Jitter gives an average variation of pitch, Shimmer gives an average variation in loudness, and the HNR gives an average variation in hoarseness.
And when loudness, pitch, and hoarseness of a voice found to be greater than a normal voice, abnormalities are detected.