Improving State-of-the-Art Continuous Speech Recognition Systems Using the N-Best Paradigm with Neural Networks