The most recommended Fakevoice technology is Coquit-AI. I downloaded this and tried running it.
First of all, I installed it to check the TTS function and voice synthesis function. I created a virtual environment Coqui without using Colab and implemented it on a local mac with git.
conda create –name Coqui python=3.8
conda activate Coqui
git clone http://github.com/coqui-ai/TTS
pip install -r requirements.txt
tts –list_models
I tried running it with one of the models. The result was saved as an ouput.mp3 file.
tts –text “What do you think you are doing? Is there any purpose to this? Sometimes I wish that I were dead” –model_name “tts_models/en/ljspeech/tacotron2-DDC”
The sound seems decent. There is also a Korean model, and I think it would be possible to make a personal model and make it with your own voice…. It seems like it’s not easy to train. For the time being, using a TTS like Sppechelo seems to be the answer.