zero-shot instruction-following

The authors introduce a multitask, multilingual, dual-modal Speech and Language Model (SLM). The SLM uses pretrained foundational speech and language models, preserving their capabilities while training a simple adapter with…