Train a DeepSpeech Model

Hey !

I’d like to train DeepSpeech 0.6.1 with this data base in order to create a new French DeepSpeech model (I know that it already exists with old versions).

Im on Windows. I downloaded Python 3.6.7 as recomended here. When Im trying to clone DeepSpeech project git clone https://github.com/mozilla/DeepSpeech and install all dependencies cd DeepSpeech pip3 install -r requirements.txt, I didn’t found requirements.txt file on the project. Is it normal ?

Please join the existing effort:

You need to checkout v0.6.1, by default you will be on master.

@kamil_BENTOUNES Please also note we have no experience training on Windows, so you might run into issues.

If you read carefully, you can see that latest 0.4 french model released does include the latest version of Common Voice data.

oh i misunderstood! I thought that v0.6.0-fr-0.4 meant French model with DeepSpeech v0.6.0. But from what you explain, it would be a model trained with version 0.4 and which could be tested with version 0.6.0 right?

I was talking about this model

okay ! It’s understood

Do you read french ? What is unclear on the release page? It’s for DeepSpeech 0.6, so it was trained with 0.6, and the 0.4 here is for the french iteration, because it adds more data compared to 0.3.4. that you link.

Yes, this one is french 0.3.4 model, that was released before the newer version of Common Voice was released. So obviously, it could not contain it.

Yes, exactly. All is more clear now