search

NVIDIA / nv-wavenet

Reference implementation of real-time autoregressive wavenet inference
BSD 3-Clause "New" or "Revised" License
735 stars 126 forks source link

how to use trained nvidia/Tacotron2 with your pre-trained nv-wavenet #67

Closed zuzana closed 6 years ago

zuzana commented 6 years ago

Hi,

I have trained nvidia/Tacotron2 model and now I'd like to use it with your pre-trained nvidia/nv-wavenet.

I'm trying to use nv-wavenet/pytorch/inferece.py. It seems thought that inference script expects model checkpoint instead of saved model.pt. (as in nv_wavenet_test.py).

When I change L42-43 in inference.py to:

 model = torch.load('model.pt')
 wavenet = nv_wavenet.NVWaveNet(**model)

I will get an error AttributeError: 'dict' object has no attribute 'get_cond_input'

Or do you have any advice how to generate cond_input.pt from nvidia/Tacotron2 so that I could use it the same way as in nv_wavenet_test.py?

Thanks.

zuzana commented 6 years ago

ok, in the end I combined this answer in nvidia/Tacotron2 repo, lines 48 - 51 from inference.py and replicating steps from get_cond_input() available in wavenet.py. generated sound is just noise but that's probably some settings.

ishansan38 commented 6 years ago

Hi @zuzana How is the progress with the integration?