Covers topics concerned with Deep Speech development
Data augmentation is important in deeplearning since it improve model invariant to some type of noises. I found that Paddle’s team do very good job how to augment speech data in ASR. Deepspeech can use these techniques to improve model performance.
We’ve currently developed an external tool to do data augmentation, the voice-corpus-tool. It allows you to do many similar things including volume, speed, noise… perturbations.