Mozilla TTS

Before Coqui, there was Mozilla TTS: a deep learning text-to-speech library that collected research-grade implementations of text-to-spectrogram models (Tacotron, Tacotron2, Glow-TTS, Speedy-Speech) and neural vocoders (MelGAN, Multiband-MelGAN, ParallelWaveGAN, WaveGrad, WaveRNN) behind a common training and inference framework, along with a GE2E-based speaker encoder and dataset quality analysis tools. Pretrained models were released in PyTorch, TensorFlow, and TFLite formats, and the stack saw use in products and research projects across more than 20 languages. Training expects a GPU while inference can run on CPU. The repository is licensed under MPL 2.0, but it is no longer where active development happens: the core team continued the codebase as Coqui TTS after leaving Mozilla. It remains relevant as a well-documented historical codebase for studying classic neural TTS architectures and for projects still pinned to its released models.