Will you release a fully trained NN?

(Rain1) #1


I see that there are now 5 datasets on this page https://voice.mozilla.org/data

but the model released here is only trained with librevox https://github.com/mozilla/DeepSpeech/releases

Will there be a release which is trained using all the datasets?

(jiping_s) #2

This issue#

says it is trained on a combination of datasets : Librispeech, Fisher and Switchboard.

(Vincent Foucault) #3

wget -O - https://github.com/mozilla/DeepSpeech/releases/download/v0.1.0/deepspeech-0.1.0-models.tar.gz | tar xvfz -
Have a nice day

(Rain1) #4

Thank you for the reply! Does this v0.1.0 have common voice trained in it?

I think this is the one I had already tried which was librevox only.

(Lissyx) #5

It was trained on more than just LibriVox as much as I recall, but it does not include Common Voice yet. We’ll do it as soon as we can :slight_smile:

(Buvana R) #6

@lissyx What are the training datasets that went into the production of the latest pre-trained model:

Was it trained on CV? And also Librivox, TED, FISHER and SWB?


(Lissyx) #7

Can you please read the replies that have been made to you ? Like here Using Deep Speech ?