open-xtts-trainer: Complete 3-Stage Training Suite for XTTS v2

Initial Release of open-xtts-trainer (v0.1.0)

This release provides the complete 3-stage training pipeline for Coqui XTTS v2:

• Stage 1 (Discrete VAE): Mel-spectrogram compression, codebook loss regularization, and support for fine-tuning from official dvae.pth.
• Stage 2 (GPT Backbone): Text tokenizer vocabulary extension for new scripts (Arabic, Devanagari, Cyrillic, etc.) and GPT fine-tuning recipe.
• Stage 3 (Latent HiFi-GAN Vocoder): Extraction of intermediate GPT conditioning latents and dual-optimizer GAN training with Multi-Period & Multi-Scale Discriminators.
• Google Colab Demo: One-click cloud GPU demonstration notebook under notebooks/open_xtts_train_demo.ipynb.

Licensed under MPL-2.0.