How to run model?

1. Download/Clone the code repository
2. Download pre-trained weights here
3. Create a folder named weights and extract content of downloaded folder there.
4. Install required packages using pip install -r requirements.txt
5. Run the command python main.py --source samples/test_sample.wav --target samples/trump10.wav. Source is the file to be converted, and target is the sample target voice.

Custom training of the voice synthesize model

Model is trained on LibriSpeech ASR corpus. Download the dataset from here

1. Encoder training

python encoder_preprocess.py <datasets_root>
python encoder_train.py my_run <datasets_root>/SV2TTS/encoder

2. Synthesizer training

python synthesizer_preprocess_audio.py <datasets_root>
python synthesizer_preprocess_embeds.py <datasets_root>/SV2TTS/synthesizer
python synthesizer_train.py my_run <datasets_root>/SV2TTS/synthesizer

3. Training the vocoder

python vocoder_preprocess.py <datasets_root>
python vocoder_train.py my_run <datasets_root>