Typically, speech models are paired with a large language model for knowledge retrieval. Block said that Dessa's knowledge stems from the conversational data it has.