《DeepSpeech Scaling up end-to-end》.pdfVIP

DeepSpeech: Scaling up end-to-end speech recognition Awni Hannun, Carl Case, Jared Casper, Bryan Catanzaro, Greg Diamos, Erich Elsen, Ryan Prenger, Sanjeev Satheesh, Shubho Sengupta, Adam Coates, Andrew Y. Ng 4 1 Baidu Research – Silicon Valley AI Lab 0 2 Abstract c e D We present a state-of-the-art speech recognition system developed using end-to- end deep learning. Our architecture is significantly simpler than traditional speech 7 systems, which rely on laboriously engineered processing pipelines; these tradi- 1 tional systems also tend to perform poorly when used in noisy environments. In contrast, our system does not need hand-designed components to model back- ] L ground noise, reverberation, or speaker variation, but instead directly learns a C function that is robust to such effects. We do not need a phoneme dictionary, . nor even the concept of a “phoneme.” Key to our approach is a well-optimized s RNN training system that uses multiple GPUs, as well as a set of novel data syn- c thesis techniques that allow us to efficiently obtain a large amount of varied data [ for training. Our system, called DeepSpeech, outperforms previously published 1 results on the widely studied Switchboard Hub5’00, achieving 16.5% error on the v full test set. Deep