kaldi parallelism

This post focus on run.pl of local parallelism.

run.pl <options> <log-file> <command>

• <options>: a job range specifier (e.g. JOB=1:4), the identifier name needn’t be JOB, and is uppercase conventionally.
• <log-file>: file name for array jobs must contain identifier of the array (e.g. foo.JOB.log)
• <command>: basically, can be anything. You need to escape pipe symbol | as \|, and other things like ; and >.

examples:

run.pl foo.log echo "hello, kaldi"

run.pl JOB=1:4 foo.JOB.log echo "hello, job: JOB"

run.pl test.log echo "foo" \| sed `s/f/F`

run.pl will replace the JOB in the command line with a number within the range.

Reference

• Parallelization in Kaldi