Blending LSTMs into CNNs
This work addresses the challenge of developing efficient and accurate speech recognition models for applications requiring real-time processing, though it is incremental as it builds on existing model compression techniques.
The paper tackled the problem of whether deep convolutional networks (CNNs) can match the accuracy of recurrent networks like LSTMs in automatic speech recognition, showing that a CNN with an image-inspired architecture outperformed previous models on the Switchboard task and further improved accuracy through model blending, achieving higher accuracy than any individual model with computational efficiency.
We consider whether deep convolutional networks (CNNs) can represent decision functions with similar accuracy as recurrent networks such as LSTMs. First, we show that a deep CNN with an architecture inspired by the models recently introduced in image recognition can yield better accuracy than previous convolutional and LSTM networks on the standard 309h Switchboard automatic speech recognition task. Then we show that even more accurate CNNs can be trained under the guidance of LSTMs using a variant of model compression, which we call model blending because the teacher and student models are similar in complexity but different in inductive bias. Blending further improves the accuracy of our CNN, yielding a computationally efficient model of accuracy higher than any of the other individual models. Examining the effect of "dark knowledge" in this model compression task, we find that less than 1% of the highest probability labels are needed for accurate model compression.