First decide what the viewer needs: captions burned into the picture for silent social viewing, a separate closed caption file a platform can switch on and off, or a fuller accessibility package with transcripts or audio description. The burned-in versus closed captions guide and the accessibility guide answer this.
Then compare auto caption tools on the languages you need, how easily you can correct mistakes and whether they export a clean SRT file. Read the edge case checklist last and run it on every file before delivery, because timing and formatting errors are easy to miss in the editor.