NVIDIA-NeMo/Speech ? reverse-engineered prompt

Build me a simple end to end speech demo using this NeMo Speech project.

I want a small app where I can upload an audio file or record from my mic, then get a transcript back on screen, and optionally turn some typed text into speech with a play button. Keep it clean and easy to run locally, and use the existing NeMo Speech code and examples if there’s already a good path for ASR and TTS. If you need the latest setup details, check the current docs online.

Please make the user experience straightforward, with a clear start page, a place to drop audio, a transcript area, and a text box for speech generation. If there are any best practices for model loading or GPU use in this repo, follow them. Also include any needed wiring, configs, and a short README note so I know how to launch it and test it.