SforAiDl/Neural-Voice-Cloning-With-Few-Samples

Implements speaker embedding extraction and multi-speaker generative modeling to achieve content-independent voice cloning with minimal samples, following the Baidu paper architecture. Combines a speaker encoder that learns voice identity fingerprints (pitch, accent, etc.) with a generative model trained on 84 speakers from VCTK dataset, enabling rapid speaker adaptation in 10-20 minutes. Built on PyTorch with DeepVoice3-inspired architecture and optimized for NVIDIA GPUs.

436 stars. No commits in the last 6 months.