What is the best practice to backup millions of voice recording with Commvault.

Jumping ahead from @Mike Struening‘s questions, since the voice recordings should stay static, I’d say use a regular file system agent in regular mode with a bunch of streams. The first backup will obviously take much longer, but incrementals should be much quicker. If you are on a more recent feature release, larger files will be backed up automatically with multiple streams (file extent backups).

Alternatively, if that first backup is simply taking too long, you can enable block level backups which skips much of the scan phase and file/open/close API which should improve performance at the sacrifice of granular restore performance (although much faster if you restore an entire disk).