Deep Recurrent Neural Networks for Acoustic Modelling
This work addresses speech recognition accuracy for ASR systems, representing an incremental architectural improvement.
The authors tackled acoustic modeling for automatic speech recognition by proposing a novel deep recurrent neural network architecture combining DNN, time convolution, and bidirectional LSTM components, achieving a 3.47% word error rate on the WSJ eval92 task with over 8% relative improvement over baseline DNN models.
We present a novel deep Recurrent Neural Network (RNN) model for acoustic modelling in Automatic Speech Recognition (ASR). We term our contribution as a TC-DNN-BLSTM-DNN model, the model combines a Deep Neural Network (DNN) with Time Convolution (TC), followed by a Bidirectional Long Short-Term Memory (BLSTM), and a final DNN. The first DNN acts as a feature processor to our model, the BLSTM then generates a context from the sequence acoustic signal, and the final DNN takes the context and models the posterior probabilities of the acoustic states. We achieve a 3.47 WER on the Wall Street Journal (WSJ) eval92 task or more than 8% relative improvement over the baseline DNN models.