Large Vocabulary Continuous Speech Recognition using RNNs

Automatic speech recognition systems are traditionally based on Hidden Markov models(HMMs) as it had given the best results in the past. However due to recent advances in neural networks, recurrent neural network(RNN) models have became competitive and surpassed HMMs in speech recognition task. RNN models also have the advantage of being used in an end-to-end model and don't require extensive hand engineering of features like HMM. In this project I aim to implement a LSTM(Long Short Term Memory) based model as based on state of the art research and make it easily available using PocketSphinx like API.