AIGC-Audio/AudioGPT ? reverse-engineered prompt

Build me a working AudioGPT demo that can handle a bunch of audio tasks from one place, like speech to text, text to speech, text to audio, sound detection, sound extraction, mono to binaural, and a basic talking head demo.

I want it to feel like a simple AI assistant for audio, where I can type a request and upload or enter the right input, then get the result back in an easy to use interface. Please wire up the existing project structure, use the pretrained models the repo expects, and make sure the main script runs cleanly with clear setup steps. If anything depends on current model docs or setup details, look them up online and keep the implementation practical.

Also, make sure the app shows helpful examples or prompts so a user can try each feature without guessing, and include sensible error messages when a model or file is missing.