DeepSpeech v0.6.0 Trained with background noise

Deep Speech
#1

I read here (https://github.com/mozilla/DeepSpeech/issues/2077#issuecomment-487451330) that there were plans to train 0.6.0 with background noise samples. Did this end up happening?

We are seeing degraded quality in noisy environments. I know this is super dependent on the type of sound, but we are considering training the model on some our own noisy samples with a super lower LR to give the model some ability to recognize our commands with noise.

Any update on the issue linked above would be great. Thanks!

#2

I think we already had enough in the release, and it was going to delay much. But @kdavis might be able to shed more light. Though, it’s holidays season, so don’t expect a reply quickly :slight_smile:

1 Like
#3

We didn’t augment the data with noise for 0.6.0.

I’d guess the first time this is going to be done is for the 1.0 release, which, hopefully, will be in the next month or two.

#4

Wow that’s soon! I look forward to that. Thank you.