Interfaze open-sources diffusion-gemma-asr-small diffusion ASR model

The model uses a diffusion decoder instead of autoregressive, supporting six languages. It is described as the first multilingual audio diffusion ASR model.