Exploring different approaches for the neural speech translation 

Abstract: 
We will talk about Deep Learning methods to perform Automatic Speech Translation (AST), e.g., cascaded as a pipeline of Automatic Speech Recognition and Text Translation, End-To-End AST models, Multi-Task and Transfer Learning techniques. We will present results of our experiments conducted on the Fisher dataset of Spanish phone conversations, discuss possible explanations for performance differences and improvements due to ensembling, beam search etc.