Status: open / Type of Theses: Bachelor Theses / Location: Dresden
Pretrained Large Language Models (LLMs) excels in natural language processing/understanding tasks by leveraging vast amounts of data collected from different sources. One particular example is text translation. Despite this prowess, LMs are biassed by the unbalanced nature of the data. In the particular case of sentence translation, they completely fail to translate popular languages such as English, French, and Deutsch to/from minority languages such as indigenous languages spoken in Africa. It is therefore crucial to reduce this balance by collecting high quality majority-minority paired data.
The aim of this project is to design and develop an innovative mobile application specifically focused on collecting spoken sentences corresponding to the translation of a sentence in French to an indigenous language spoken in Africa. The app will provide users with a series of French sentences displayed on the screen, which they will translate to their mother language, and record the pronunciation with the app. The collected data will contribute to the enhancement of speech recognition systems and language processing technologies.
By undertaking this project, not only will you gain valuable experience in designing and developing mobile applications but also contribute towards advancing speech recognition systems using real-world data collected from native speakers. If you are interested in pursuing this exciting opportunity as your bachelor thesis project or a research project, please contact Dr. Kossi Amouzouvi at kossi.amouzouvi@tu-dresden.de to discuss further details.