erew123/alltalk_tts ? reverse-engineered prompt

Build me a local text to speech app called AllTalk that I can run either by itself or alongside other tools like text generation apps. It should let me type in text, pick or upload voices, and generate clean audio files quickly. I want a simple web page for settings and documentation, plus an API I can call with JSON from other apps. Please include support for low VRAM use, faster generation when possible, model downloads, custom local models, and a finetuning workflow for training on a specific voice. It should also have a bulk generator for longer scripts, a narrator mode for using different voices for dialogue and narration, and an option to clean up old wav files automatically. Make the setup easy on Windows and Linux, with clear console messages and a straightforward launch flow. If you need to check current docs for the TTS engine or related tools, go ahead and look them up online.