Zohre Foroushi defends her PhD thesis

Zohre successfully defended her PhD thesis:

Audio-Visual Speech Enhancement using Unsupervised Learning Methods.

This PhD thesis develops a novel unsupervised audio‑visual speech enhancement framework that combines sound and lip‑movement cues to improve speech clarity in noisy environments. It introduces an audio‑visual recurrent variational autoencoder enhanced with advanced probabilistic sampling, a silence‑aware noise model, and a new video inpainting method to handle partially occluded faces. Together, these innovations significantly improve speech quality and intelligibility—especially in challenging real‑world conditions—while demonstrating strong performance across multiple benchmark datasets.

Congratulations !