Semi supervised accent invariant speech recognition

Abstract

Automatic Speech Recognition which is aimed at enabling a more natural form of human-machine interaction has been an area of research for decades now and many breakthroughs have been made in this eld. The performance of many state-of-the-art systems for mainstream languages and accents are extremely good. And there is a need for zero-resource or minimal resource systems as gathering enough data is highly challenging and sometimes near impossible. Through this study, we aim to provide a proof of concept for the idea of using speech embeddings for Automatic Speech Recognition.