TTS-experiments Probably nothing here First idea to try is to train a vocoder on the output of a text encoder with a single speaker dataset (e.g. LJSpeech)