Shua’s research centers on unsupervised and self-supervised learning methods for speech recognition, speaker verification, and diarization, particularly under challenging conditions. In his MSc, he developed the first deep learning algorithm capable of accurately estimating and tracking formants.

Roni’s work is focused on both the security of deep learning systems in general and of speech application in particular. She is also interested in unsupervised and self-supervised representation of the speech signal and its applications to automatic analysis of linguistic data.

Eyal’s research explores the integration of large language models into speech recognition and processing systems. During his MSc studies, he developed a method for time-scale modification of speech signals using GANs, enabling more natural and flexible manipulation of audio timing without compromising quality.

Roi specializes in Diffusion Models for speech and sequential signals, text-to-speech applications using diffusion models, and the spectral analysis of these models.

Development of new algorithms for accurate phoneme and word forced alignment.

Investigating the alignment between brain activity (measured via fMRI) and internal representations of the Whisper transformer-based speech recognition model.

Transforming transformer-based speech recognition models (e.g., Whisper) into efficient streaming systems.

Explainability and interpretability of transformer-based speech models.

Enhancing corrupted signals using diffusion models to improve classification performance.

Reasoning in speech language models.

Cross-domain speaker verification using diffusion models.