If you have byte arrays, you must first convert those byte arrays to the correct sample format, such as 16-bit integer (short). Once you've done that, you have a few options, including:
Average the data, sample-by-sample.
Sum the data, sample-by-sample, and then "clip" any samples that are out of range.
The first option is the only correct way to do it without distorting your signal.
Finally, you may need to convert back to bytes to save to a file.
For more details, including code, you may want to see something more like a tutorial.