Method combines text LLM weights to create instruction-following speech model without costly tuning

Original: Unlocking Speech-Text Compositional Powers: Instruction-Following Speech Language Models without Instruction Tuning

Source: arxiv.org ↗