Text to Speech
During last weeks I’ve been playing with TTS (Text-To-Speech) … looking for Bit Robot (Inmoov) voice. I trained WaveRNN from scratch using LJSpeech dataset and after I trained Tacotron and Forward-Tacotron. It took some GPU-days and even it’s my first tests, results are pretty good. Now training bit more to use with MelGAN.
You can find Forward-Tacotron code here
Soon I will publish “behind the scenes” post on Patreon about how to setup and train the system and provide access to my fork with trained models and some fixes because some requirements are broken. So don’t forget to check.