225 Cepstral and fourier based methods are feasible feature extraction approaches for parkinsons disease voice analysis

Introduction The National Institute of Healthcare Excellence estimates that each year, around 17,300 new cases of Parkinson’s disease are diagnosed in individuals aged 45 and above, translating to an incidence rate of 33.4 per 100,000 person-years. The application of artificial intelligence to the classification of on-off speech patterns in Parkinson’s disease is being revolutionized through the integration of advanced feature extraction techniques such as Mel-Frequency Cepstral Coefficients (MFCCs), Fourier Transform, and wavelet analysis. These methods enable the capture of the short-term power spectrum, frequency content, and localized time-frequency representations of speech signals, effectively mirroring human auditory perception and highlighting subtle differences in vocal patterns.Methods MFCCs, in particular, excel at modeling human auditory processing by emphasizing key speech characteristics, while Fourier Transform and wavelet analysis provide complementary insights into spectral and temporal patterns. Advanced machine learning algorithms analyze these features to distinguish between ‘on’ periods—when medication yields improved motor control—and ‘off’ periods, marked by speech impairment. Attributes such as rhythm, prosody, articulation clarity, and voice intensity are extracted and analyzed to track nuanced transitions between motor states in real time.Discussion Empirical validation on 200 audio data of Parkinson’s patients’ speech demonstrates that these models achieve high specificity of 100%. This multi-technique integration ensemble artificial intelligence approach outperforms state of the art support vector machines and could enhance diagnostic precision, informing treatment optimization, and providing future valuable options for clinicians.jdavids@ic.ac.uk