Analysing Audio-visual Self-supervised Learned Speech Representations

My focus was on the application of audio-visual SSL representations in low-resourced speech recognition, for example, testing their performance when trained with multilingual data and fine-tuned on lower-resourced data. This year, my research trajectory was inspired by some studies on the audio-visual integration that link audio-visual SSL models to a cognitive science perspective. Consequently, I shifted my focus of analysis on the audio-visual SSL speech representations to their temporal dynamic of phonetic encoding and dropped my target analysing from the aspect of cross-lingual transfer.