Combining Frame-Synchronous and Label-Synchronous Systems for Speech Recognition