Abstract: Synchronizing data with metadata includes simultaneously triggering a metadata generation unit to produce metadata that are stored in a reliable metadata queue, and a data unit producing unit that produces data units that are processed in a lossy processing pipeline. A synchronizer obtains a processed data unit, and considers one or more statistics associated with the metadata queue to determine whether it is likely that a data unit was dropped by the lossy processing pipeline. For example, an increase in the depth of the queue may be used to infer that a data unit was dropped. If so, enough metadata entries may be dequeued such that the next metadata entry is correctly synchronized with the data unit. The synchronized data unit and metadata unit are output together. Another technique may take the output from two or more different systems and align data units from each system based on the metadata.