alltalk_tts

Why we included this project

AllTalk is aimed at people who already run a local text generation setup and want it to speak. It works standalone or slots into tools like SillyTavern and the oobabooga web UI, so it fits roleplay sessions and chatbot demos that need narration. The practical part is the tuning: low-VRAM mode keeps it running alongside an LLM that has eaten most of your GPU memory, DeepSpeed speeds up generation, and bulk mode reads long documents into a single audio file. You can fine-tune a model on a specific voice or load custom XTTSv2 voices, which helps when you need a consistent character voice instead of the default one. There is also a JSON API for driving it from your own scripts, and the built-in documentation and settings page make the local-first setup easier to manage.