We contacted the British Broadcasting Corporation (BBC)…

The datasets we hoped to use to capture a wide variety of speakers while still maintaining the accuracy of the machine learning model are LRW and LRS2, but we hit a stop: We need the permission of the British Broadcasting Corporation to gain access to the data. Hence, we have contacted them for access and I am currently waiting on their reply. The feeling to research something so highly advanced, such as audio-visual speech recognition, with professional data is very exciting for me! This will definitely serve as a great research area for me in the future as a college student and beyond because I really love understanding and dealing with the complexities of machine learning models to optimize an effective solution. In the meantime, we have discovered two possible backup datasets that we could use, which are precisely the Grid Dataset and an AVSR dataset.

Now, I hope to look at the machine learning model and the specific code implementation rather than the theory in the research papers that I already explored in the past few weeks. I will try to understand the various parts of the layers in the machine learning model and preprocessing steps in order to learn new coding implementations with new libraries, methods, or algorithms that can help build my machine learning model seamlessly.

So see you next week, same place, same time.