How to clone my own voice locally

[gitihobo]

I want to create my own dataset, how do I do this? I cant seem to make sense of the dataset section

[bitplane]

Dunno if it'll work but I did this:

• Used Mozilla Common Voice to help build a dataset for the whole world
• Downloaded my data containing a couple of thousand recordings through the Common Voice export page in my profile
• Converted the mp3 files in the zip to wav files and put them in a folder called wavs
• Then took the text from the export and put the name of the wav followed by the text, pipe separated: wav_file.wav|Text spoken, and put it in metadata.csv
• replaced all the fancy quotes for " and the fancy apostrophes with '
• zipped this up
• Ran docker build . -t voice-cloning-app in the project dir
• Then docker run --gpus all -p 5000:5000 -v$(pwd)/data:/app/data voice-cloning-app
• Then went to http://localhost:5000/ and uploaded the zip file as an import
• It then said it'd take a year to run, so I exported it and used the colab notebook. Had to make a couple of changes to get it working:

https://colab.research.google.com/gist/bitplane/60543855a8fc3df473e94e18d45b06a3/copy-of-voice-cloning-training.ipynb

You'll need this file in your Drive dir too, call it pretrained.pt:

https://drive.usercontent.google.com/download?id=1c5ZTuT7J08wLUoVZ2KkUs_VdZuJ86ZqA