Deep Speech: Scaling up end-to-end speech recognition
At a glance
A neural network trained on thousands of hours of audio simplified speech recognition and improved performance in noise.
In one paragraph
Introduced a speech recognition system that learned to turn audio into text, replacing many hand-built processing steps with a neural network. Training on thousands of hours of speech using multiple GPUs helped it recognize speech in noisy recordings.
Authors 11
Organizations
Credited for this work at publication.