Aratako/Irodori-TTS ? reverse-engineered prompt

Build me a text to speech app that can clone a voice from a reference audio clip and generate a new spoken version of my text in that same voice.

I want it to support Japanese text, let me optionally add a short caption to control the speaking style or emotion, and also handle emoji in the input text to change the delivery when possible. It should work with one or more reference audio files, and if I do not provide a reference, it should still generate speech from text alone. Please include a simple command line way to run inference and a small web UI where I can paste text, upload audio, and download the finished wav file.

If you need to, look up the current docs online for the model and any setup details. Make the whole thing easy to run locally and include a clean README with example commands.