lmnt-com/diffwave

Based on diffusion probabilistic models, DiffWave iteratively refines Gaussian noise into audio conditioned on mel-spectrograms or synthesizes unconditionally. It supports mixed-precision and multi-GPU training via PyTorch, with a fast sampling algorithm achieving 0.87 real-time factor on pretrained models. Inference is available through both a Python API and command-line interface for integration into speech synthesis pipelines.

885 stars. No commits in the last 6 months.