This technology is a pitch-synchronous voice analysis method that separates timbre from pitch to generate more precise vocal biomarkers for non-invasive disease screening and monitoring.
Voice-based diagnostic methods commonly rely on acoustic features such as Mel-frequency cepstral coefficients (MFCCs), which are calculated over fixed time windows. This approach can mix pitch and timbre information and miss subtle changes within individual voice cycles, reducing the sensitivity and reproducibility of resulting biomarkers. More precise acoustic features are needed to support reliable comparison across voice samples and improve the performance of diagnostic models.
This technology is a pitch-synchronous voice analysis method that analyzes voice signals in alignment with individual pitch periods rather than using fixed time windows. It extracts pitch-synchronous timbre vectors and instantaneous pitch information, allowing pitch and timbre characteristics to be evaluated separately while preserving fine-grained acoustic variation. The resulting features can be compared using a timbre-distance metric and used as biomarkers in voice-based screening, monitoring, and classification models.
Demonstrated in silico with simulations and initial analyses showing enhanced feature extraction compared to traditional MFCCs.
Patent Pending
IR CU26053
Licensing Contact: Dovina Qu
