The recent advancements in speech processing research are significantly pushing the boundaries of what is possible in the field. A notable trend is the increasing use of transformer-based models, which are demonstrating superior performance in tasks such as distinguishing scripted from spontaneous speech across multiple languages. These models are not only advancing the state-of-the-art but also highlighting the potential for generalisation across different formats and languages. Another significant development is the focus on efficient and high-quality data collection, leveraging Speech Foundation Models to automate validation processes, thereby reducing costs and improving scalability. This approach is particularly promising for multilingual contexts, as evidenced by studies in dysarthric speech assessment and target speaker extraction. Additionally, there is a growing emphasis on creating and utilizing large, diverse datasets, such as the Libri2Vox dataset, which combines real-world and synthetic data to enhance model robustness. These advancements collectively suggest a shift towards more automated, scalable, and multilingual solutions in speech processing, with a strong emphasis on leveraging cutting-edge machine learning techniques and comprehensive datasets.