Provided are methods for synchronizing multiple video recordings of the same event using audio signals of the recordings. An algorithm is designed to generate candidate time offsets for pairwise matches of video recordings based on audio signals corresponding to the recordings, and combine the candidate time offsets to obtain the most accurate actual time offset between each pair of recordings. Each candidate time offset for each pairwise match of recordings is assigned a confidence measure indicating how likely it is that the particular time offset is correct for the given pair of recordings. The confidence measure assigned to a particular time offset is based on various components and characteristics of the audio signal pair corresponding to the video recordings.