On the Compression of Recurrent Neural Networks with an Application to LVCSR acoustic modeling for Embedded Speech Recognition
This enables more compact and accurate speech recognition systems for mobile devices, but it is incremental as it builds on existing compression techniques.
The paper tackles the problem of compressing recurrent neural networks, specifically for acoustic models in speech recognition, and achieves a reduction in model size to one-third of the original with negligible accuracy loss.
We study the problem of compressing recurrent neural networks (RNNs). In particular, we focus on the compression of RNN acoustic models, which are motivated by the goal of building compact and accurate speech recognition systems which can be run efficiently on mobile devices. In this work, we present a technique for general recurrent model compression that jointly compresses both recurrent and non-recurrent inter-layer weight matrices. We find that the proposed technique allows us to reduce the size of our Long Short-Term Memory (LSTM) acoustic model to a third of its original size with negligible loss in accuracy.