TICL: Text-Embedding KNN For Speech In-Context Learning Unlocks Speech Recognition Abilities of Large Multimodal Models Paper • 2509.13395 • Published Sep 16 • 1