Intelligent Speech Analysis For Facial Paralysis: An Innovative Diagnostic Approach

Facial paralysis is a significant focus in oral and maxillofacial surgery. Alterations in the acoustic characteristics of speech stemming from facial nerve dysfunction provide new biomarkers for the objective evaluation of facial paralysis. However, there has been little research attention so far. In this ethically approved study, we recruited 100 patients (aged 18-70) with idiopathic facial nerve paralysis, alongside a cohort of age- and gender-matched healthy control group. We collected patients’speech data using the tailored facial paralysis speech assessment scale and standardized speech acquisition device. This study pioneered intelligent speech analysis technology to propose a Dynamic Acoustic-based Facial paralysis diagnosis Early Screening & Treatment System (DAFESTS), which overcomes the subjectivity inherent in conventional visual appraisals like House-Brackmann Grading while offering a more cost-effective alternative to electromyography. With DAFESTS, we constructed a six-dimensional acoustic feature matrix, including fundamental frequency, nasalization degree, formant frequencies, speech tempo, pause dynamics, and spectrographic signatures, and developed an emerging phoneme sensitivity hierarchical analysis model grounded in phonetic principles. This research established the first database of speech biomarkers for facial paralysis. Moreover, the developed sensitivity feature extraction algorithm in DAFESTS enables the identification of patients exhibiting atypical early clinical manifestations, facilitates the prognostication of facial nerve injury risk, tracks disease trajectory, and quantifies the severity. Clinically, this research furnishes quantitative, interpretable AI-driven insights for the diagnosis and therapeutic management of facial paralysis. Concurrently, it provides evidencebased foundations to guide the evaluation of treatment outcomes.