Luise - DREU Research Intern
Here at Columbia, I work on the project “Text-to-Speech Synthesis for Low-Resource Languages” with the Spoken Language Processing Group. We are working towards developing methods that will allow us and others to develop intelligible and natural sounding text to speech (TTS) technologies from limited data.
While the end goal is concerned with TTS, I mainly work with automatic speech recognition (ASR). The data we work with is often not just limited but also noisy (e.g. from phone conversations), so we want to select the data that gives us the best TTS. The idea is that data that gives good ASR results also will give good TTS results. It is my job to train models on datasets of different types and sizes to find the best combination.
You can read more about the projet here.