Speech Segmentation with Neural Networks to Facilitate Vernacular Translation

Abstract

The concept of using Artificial Intelligence (AI) in improving the lifestyle of society emphasizes on the fact that it should reach and benefit every community; In compliance to this, overcoming language barriers is still prevalent in many countries. Kannada is the spoken language of more than 50 million people in the world (2011). However, dialects of Kannada are diverse and vary according to geographic distributions making current machine translation research work difficult and less attended to. This research ventures to construct English – Kannada neural machine translation systems using Machine Learning techniques. The Corpus in our Model is trained using Kannada & Coastal Kannada dictionaries, Wikipedia & news articles. In this paper, we’ve implemented sequence-to-sequence learning using encoder-decoder LSTM (Long Short-Term Memory) model. We also review the significance of the usage of attention mechanism as it is an added strategy to improve the performance of the system. Developing a translation system to overcome linguistic barriers is the main goal of this research.