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 systems, which rely on laboriously engineered processing pipelines; these traditional systems also tend to perform poorly when used in noisy environments. In contrast, our system does not need hand-designed components to model background noise, reverberation, or speaker variation, but instead directly learns a 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 RNN training system that uses multiple GPUs, as well as a set of novel data synthesis techniques that allow us to efficiently obtain a large amount of varied data for training. Our system, called DeepSpeech, outperforms previously published results on the widely studied Switchboard Hub5’00, achieving 16.5% error on the full test set. DeepSpeech also handles challenging noisy environments better than widely used, state-of-the-art commercial speech systems.
Although the academic papers, so far, are using “deep learning” in a meaningful sense, early 2015 is likely to see many vendors rebranding their offerings as incorporating or being based on deep learning.
When approached with any “deep learning” application or service, check out the Internet Archive WayBack Machine to see how they were marketing their software/service before “deep learning” became popular.
Is there a GPU-powered box in your future?
I first saw this in a tweet by Andrew Ng.
Update: After posting I encountered: Baidu claims deep learning breakthrough with Deep Speech by Derrick Harris. Talks to Andrew Ng, great write-up.