Pretrained model using all three corpus
shiinoandra opened this issue · comments
First, I would say thank you for making this awesome repository,
I've been trying to train the model with all the dataset from 3 corpus but it seems will take really long time (need about 30 hours for single epoch. I'm using GTX1070 in SLI).
is it possible to provide a pretrained model with all the three dataset? as the current pretrained model is trained only with VCTK dataset if I'm not wrong
Thank you very much
silly me, i haven't configure the CUDA for the tensorflow in the ubuntu no wonder it runs really slow. after i built again from the source it now detect the CUDA lib and the training is running at decent speed. Training the model using all the corpus now, will report back later
Is it possible for you to share the pretrained model which was obtained for all 3 corpuses?
Thanks
I'm 40 epochs in to the training with a 1080ti. Will share the entire training folder output when it's done with 50.
entire contents of my /train/ directory (4.5GB zipped) : https://ufile.io/mivf6
output from the newly trained model using the same examples...
he hoped there would be stoe for dinner turnips and carrits and brose patitoes and fatin lutton pieces to be latled out an thick peppered flower faten sauce
stuffed into you his belly countled him
after early night fall the yellowe lamps whuld light hup pher and there the squalid quarter of the browfls
ilo burti and e good in your mind
nomebut dan freshn knawly is waiting on you could night husband
original example results for comparison...
he hoped there would be stoo for dinner turnips and charrats and bruzed patatos and fat mutton pieces to be ladled out in th thick peppered flower fatan sauce
stuffid into you his belly counsiled him
after early night fall the yetl lampse woich light hop here and there on the squalled quarter of the browfles
o berty and he god in your mind
numbrt tan fresh nalli is waiting on nou cold nit husband
@greigs could you renew your uploaded file or put it on a public server somewhere? ufile.io is pure dark pattern (clickbait, ads, data hostage taking). That isn't the sort of place you'd like to share your hard-won model data. I'd be happy to host your 5GB model, if you could figure out a way to get it to me (dropbox, S3, google drive?).
@greigs Do you still have the pretrained model?
@AwnishIndus @ebadali @hobson Here is the file uploaded by @greigs. https://drive.google.com/open?id=1RpAg48ZFIsPWm3dC5GU34pRWEXkOH1S- I reuploaded it to google drive.
Hi @greigs , @MarinescuEvghenii , What is your CTC loss of using all three corpus. I use all three corpus and I got about 103 CTC loss in Epoch 6. After that, the CTC loss became inf. Thanks.