TTS

Why we included this project

For anyone who wants to run neural text-to-speech on their own hardware, this PyTorch project bundles the pieces that usually have to be assembled by hand: text-to-spectrogram models like Tacotron 2 and Glow-TTS, a speaker encoder, and vocoders such as MelGAN and WaveGrad. Pretrained models cover 20+ languages, so you can go from install to speech quickly, and the dataset-analysis tools plus a flexible dataloader make it practical to retrain on your own recordings when a stock voice doesn't fit. It also supports multi-speaker training, multi-GPU runs, and export to TensorFlow/TFLite for on-device inference, which matters for teams shipping voice assistants or accessibility features. If you're evaluating neural TTS and want one codebase to experiment in, this is a strong candidate.