drama-3-tr: Fish Audio Drama 3 in Turkish on one 16 GB GPU

About

drama-3-tr runs the Fish Audio Drama 3 text-to-speech model in Turkish on a single 16 GB GPU. The full model reads the prompt, a distilled 32-expert speaker generates the audio, and expert weights are streamed from an NVMe drive so the 41.5 GB checkpoint does not need to fit in RAM. The README reports about 2.7 times real time on short sentences and 3.3 times on paragraphs, with inline emotion directions preserved.