It is overly verbose and easy to make mistake.
If user can use CSV file, they can easily use any spreadsheet on the market to create a parallel corpus.
Create Python app that take glob pattern and perform batch conversion of the files.
The tool is in wlt-dataset sub-project. There're two files, csv2tmx.py and tmx2csv.py to do the job.
It is overly verbose and easy to make mistake. If user can use CSV file, they can easily use any spreadsheet on the market to create a parallel corpus.