Probably something like https://github.com/openai/whisper which isn’t an LLM, but is a different type of model dedicated to speech recognition