deterministic-algorithms-lab/Cross-Lingual-Voice-Cloning

Extends Tacotron 2 with multilingual and multi-speaker conditioning via speaker and language embeddings, enabling synthesis across language pairs using a single trained model. Supports distributed training with automatic mixed precision via NVIDIA Apex, and accepts datasets in standardized format with speaker/language ID annotations per utterance. Mel-spectrogram generation can be paired with vocoder backends like WaveGlow for audio synthesis.

360 stars. No commits in the last 6 months.