search

CorentinJ / Real-Time-Voice-Cloning

Clone a voice in 5 seconds to generate arbitrary speech in real-time
Other
52.3k stars 8.75k forks source link

How to use single speaker trained model for voice cloning? #996

Open thasin96 opened 2 years ago

thasin96 commented 2 years ago

I have tried this link to create sample dataset of one speaker. From the dataset, performed encoding and training synthesizer and Vocoder model training. After the models have been trained I got model files inside saved_models/<experiment>.

Then, I wanted to use the custom trained model instead of default model. So, tried this command python demo_toolbox.py -m ./saved_models/my_run/. Instead of taking the custom models, it downloads default model again and inferencing based on the default model.

How to properly use trained model for producing inference? Any help would be appreciable. If you need more detailed, please let me know.

Korean88 commented 1 year ago

I'm not sure if this is still relevant. If it is, please take a look at line 34 of demo_toolbox.py. If you open utils.default_models.py you'll see the method checks the size of the models and downloads the default ones from goodle drive if the size is different from the expected. So you can try to remove line demo_toolbox:34 or comment it out