Turn-taking and Backchannel Prediction with Acoustic and Large Language Model Fusion