Realistic text-to-speech open source libraries

[sacred mirage]

Hey all! I would appreciate some suggestions for:

1- The best text-to-speech open source library at the moment that could get a quality close to what PLAY.HT has.
2- Could be trained to speak in your own voice.

Thanks.

[prime zephyr]

There are a ton of TTS models available. Look up FastPitch, Flowtron, FastSpeech2, Grad-TTS, Mixer-TTS, TalkNet. Many of these you can find on Nvidia repos:
https://github.com/nvidia-riva
https://github.com/NVIDIA/NeMo
https://github.com/NVIDIA/flowtron
https://github.com/NVIDIA/DeepLearningExamples/tree/master/PyTorch/SpeechSynthesis
https://github.com/huawei-noah/Speech-Backbones/tree/main/Grad-TTS

Each of these has been fine tuned on LJSpeech so there's a pre-trained model to download. What I would do in your position is create several samples of yourself to fine-tune the model on and resume the training now using only your audio data.