This repo is a WIP copy of my Master's Thesis at the Institute for Anthropomatics at the Karlsruhe Institute of Technology at the "Multilingual Automatic Speech Recognition" Research Group, written in the 6 months from December 2016 to May 2017.
It includes an approach to identify languages based on Bottleneck Features of each frame produced from a normal Automatic Speech Recognizer trained on 10 languages.
Training was completed using DNNs as well as RNNs and post-smoothing using a variety of filters.