Efficiently managing the radio spectrum in 6G to simultaneously serve both terrestrial and non-terrestrial networks (NTNs) is critical for realizing the 6G vision. However, the diversity of deployment scenarios and dynamic nature of traffic loads make it challenging through model-based optimization approaches. This paper proposes a machine learning based framework for intelligent and dynamic spectrum allocation between terrestrial and NTN links in 6G systems. A spectrum controller gathers data about channel conditions, user demands, network traffic, etc., and trains a neural network model to learn complex inter-dependencies. The proposed framework provides an intelligent spectrum management solution for integrated terrestrial-NTN networks in 6G.