Advances in Joint CTC-Attention based End-to-End Speech Recognition with a Deep CNN Encoder and RNN-LM
This work improves speech recognition accuracy for spontaneous speech in Japanese and Chinese, though it appears incremental as it builds on existing joint CTC-attention approaches.
The authors tackled automatic speech recognition by developing an end-to-end model combining CTC and attention mechanisms with a deep CNN encoder and RNN language model, achieving a 5-10% error reduction on spontaneous Japanese and Chinese speech compared to prior systems and outperforming traditional hybrid ASR systems.
We present a state-of-the-art end-to-end Automatic Speech Recognition (ASR) model. We learn to listen and write characters with a joint Connectionist Temporal Classification (CTC) and attention-based encoder-decoder network. The encoder is a deep Convolutional Neural Network (CNN) based on the VGG network. The CTC network sits on top of the encoder and is jointly trained with the attention-based decoder. During the beam search process, we combine the CTC predictions, the attention-based decoder predictions and a separately trained LSTM language model. We achieve a 5-10\% error reduction compared to prior systems on spontaneous Japanese and Chinese speech, and our end-to-end model beats out traditional hybrid ASR systems.