PaddlePaddle/Parakeet

Built on PaddlePaddle's dynamic graph framework, Parakeet implements a modular two-stage TTS pipeline combining acoustic models (mel-spectrogram prediction) with neural vocoders for waveform synthesis. It provides standardized data preprocessing, text frontend processing (including rule-based Chinese text analysis), and enables voice cloning through speaker embedding transfer learning. The toolkit targets multilingual synthesis across Chinese and English datasets (CSMSC, AISHELL-3, LJSpeech) with pre-trained checkpoints for production deployment.

618 stars. No commits in the last 6 months.