939 B
voice-cloning
Version updates
2022.05.19： I calculated GE2E loss in encoder with cuda rather than originally-configured CPU. It speeds up the encoder training speed.
2022.07.15： I added Loss animation plot for synthesizer and vocoder.
2022.07.19： I added response time and Griffin-Lim vocoder results for demo_toolbox.
2022.07.29： I added model validation for encoder, synthesizer and vocoder.
2022.08.02： I added voxceleb train and dev data for encoder. I added noise reduce method for the output wav from vocoder.
noisereduce reference: https://github.com/timsainb/noisereduce
2022.08.06： I split the long text into short sentences using spacy for input of synthesizer. Make sure to install English dataset en_core_web_sm, say by python -m spacy download en_core_web_sm
2022.09.02： I set prop_decrease=0.6 for male and 0.9 for female in noisereduce function