deepvoice3_pytorch

Why we included this project

DeepVoice3_pytorch is a PyTorch port of the convolutional text-to-speech model from the 2017 Deep Voice 3 paper, the fully convolutional sequence-to-sequence architecture with attention. The repo pairs it with the related guided-attention variant and includes preprocessors for LJSpeech, JSUT, and VCTK, plus a custom JSON format, along with English and Japanese frontend text processors, so you can train single or multi speaker models yourself. Pre-trained models and audio samples let you judge output quality before spending time on training. The codebase is from 2017 and needs some dependency and PyTorch version cleanup, but for studying how these convolutional TTS systems are wired together, or reproducing their results, it remains a worthwhile reference.