CNN-based MultiChannel End-to-End Speech Recognition for everyday home environments
This work addresses speech recognition challenges in noisy home environments, but it is incremental as it builds on existing end-to-end and multichannel methods.
This study tackled the problem of automatic speech recognition in noisy everyday home environments with multiple speakers, using a CNN-based multichannel end-to-end system, and achieved a word error rate reduction of 8.5% absolute from a single-channel end-to-end baseline and 0.6% from the best baseline on the CHiME-5 corpus.
Casual conversations involving multiple speakers and noises from surrounding devices are common in everyday environments, which degrades the performances of automatic speech recognition systems. These challenging characteristics of environments are the target of the CHiME-5 challenge. By employing a convolutional neural network (CNN)-based multichannel end-to-end speech recognition system, this study attempts to overcome the presents difficulties in everyday environments. The system comprises of an attention-based encoder-decoder neural network that directly generates a text as an output from a sound input. The multichannel CNN encoder, which uses residual connections and batch renormalization, is trained with augmented data, including white noise injection. The experimental results show that the word error rate is reduced by 8.5% and 0.6% absolute from a single channel end-to-end and the best baseline (LF-MMI TDNN) on the CHiME-5 corpus, respectively.