search

SeanNaren / deepspeech.torch

Speech Recognition using DeepSpeech2 network and the CTC activation function.
MIT License
260 stars 73 forks source link

Trained model file download #70

Closed zavalyshyn closed 7 years ago

zavalyshyn commented 7 years ago

Hi,

I've been running the Train.lua script for days on my setup (no GPU, just CPU) and managed to move to the second epoch.

Is there a way to download a pretrained deepspeech AN4 model from somewhere?

I'd really appreciate this. Thanks

SeanNaren commented 7 years ago

Hey sorry for the late response. I'll do this when I can, I plan to release two models for Librispeech and for AN4, but not entirely sure where to put it, any suggestions?

EDIT: I'll make a release and put it there

zavalyshyn commented 7 years ago

Hey, cool thanks.

I can store it on my Dropbox. I have 27 gigs there which I don't use.

shantanudev commented 7 years ago

You can also store it on my google drive if you are interested. I have unlimited space.

nn-learner commented 7 years ago

@SeanNaren Thank you Sean! What is the best WER% you have been able to achieve so far?

SeanNaren commented 7 years ago

Sorry for the disappearance, everything can be seen here!

SeanNaren commented 7 years ago

I realised some people may want a CPU only model (would take forever to train on Libri, I can do this for AN4 only). I'll make a ticket for this.

zavalyshyn commented 7 years ago

Thank you so much for that. That's exactly my case.

On Thu, Jan 5, 2017 at 3:21 PM, Sean Naren notifications@github.com wrote:

I realised some people may want a CPU only model (would take forever to train on Libri, I can do this for AN4 only). I'll make a ticket for this.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/SeanNaren/deepspeech.torch/issues/70#issuecomment-270669370, or mute the thread https://github.com/notifications/unsubscribe-auth/AG3PfAIjxxg2w3Osj_OHmDCJzrlxgkosks5rPQqUgaJpZM4K5XOD .