Voice technologies for health

AI-driven voice recognition, voice quality assessment, speech enhancement, and speech generation—with a distinctive focus on multi-objective optimization to balance accuracy, robustness, and clinical usability.

• Voice reconstruction & speech synthesis for alaryngeal patients: flow-based generative models (e.g., P-GLOW) and clinical speech replacement/enhancement systems (SpeechEnhancer) aimed at post-laryngeal-surgery voice restoration.

• Clinical voice assessment at scale: Pareto-optimized AVQI assessment with device-agnostic scoring, validated workflows, and cross-smartphone consistency.

• Neurodegenerative screening from voice: hybrid deep learning (e.g., “U-lossian” networks) for Parkinson’s disease detection, designed for heterogeneous speech tasks and strong PD/non-PD discrimination.

• Robust speech enhancement in noise: specialized models such as Pareto Denoising Gated LSTM (PD-GLSTM) to reduce artifacts and handle unvoiced frames in challenging speech (including tracheoesophageal/alaryngeal speech).

• Signal-processing + ML pipelines for impaired speech cleaning: Pareto-Optimized NMF (PONNMF) with objective evaluation using AVE and AI-based ASVI indices.

• Translational mobile health tooling: Voice Wellness Index (VWI) app (iOS/Android) combining AVQI + GFI with server-side processing, enabling screening-grade performance and telemedicine-ready deployment.