← Lectures·L8

Multimodal Analysis of Human Phonation

Kniesburges · 1 concepts · 7 questions

Concept 1 / 1

Key Concepts to Memorize

Physiology of phonation:


  • Vocal folds = primary sound generator; ventricular folds are passive
  • Larynx serves two functions: (1) swallowing (3-step valve), (2) speech/singing
  • Fundamental frequency of voice: f0 = 150–1500 Hz
  • Physics of phonation = Fluid-Structure-Acoustic Interaction (FSAI)

Dysphonia = voice disorder:


  • Symptoms: hoarseness, decreased load capacity, incomplete glottis closure, asymmetric oscillations
  • Organic Dysphonia (structural cause): malformation, trauma, inflammation, malignant/benign growth (e.g., polyp, squamous cell carcinoma)
  • Functional Dysphonia (no primary organ findings): over/incorrect loading, multiple combined causes

Clinical diagnostics of dysphonia (multimodal):


  • 2D visualization: Laryngoscopy, Stroboscopy, Highspeed endoscopy
  • Acoustic signal analysis: Voice field measurement, irregularity parameters
  • Self + expert evaluation
  • ElectroGlottoGraphy (EGG)
  • Key limitation: in vivo examination of sound generation during phonation is NOT completely possible

Deep learning in laryngoscopy (3 tasks):


1. Localization of the glottis and vocal folds

2. Automatic segmentation of the glottis area

3. Classification of tissue type, organic disorder, etc.


  • BAGLES benchmark: 7 hospitals (EU + US), 640 records, 5 cameras, 59,250 images with segmentation

3 types of larynx models:

Model TypeDegree of RealityAI SupportData Density
Ex vivoHighestSomeMedium
Synthetic (silicone)MediumMediumMedium
Computational (CFD)LowestHighestHighest

AI-supported CFD simulations:


  • Classical CFD = extremely slow: 140 cores, 10h per cycle → 100h for 10 cycles
  • Solution: SIREN (Implicit Neural Representations with periodic activation functions)
  • SIREN enables: (1) increase spatial resolution, (2) increase temporal resolution, (3) future prediction of flow fields

Take-home message: AI in biomedical science goes far beyond MRI/CT postprocessing.