AugSumm: Towards Generalizable Speech Summarization Using Synthetic Labels from Large Language Models | Synapse