xiaomi-research / midashenglm-gen

git clone https://github.com/xiaomi-research/midashenglm-gen.git

LLM‑driven autoregressive flow matching for unified speech‑music‑sound‑effect audio scene generation