Voice dubbing is where a client wants their content translated in a foreign language but instead of through written subtitles, they desire an actual spoken audio track. The process of voice dubbing is infinitely more time-consuming and difficult than it may appear. On average it takes one hour of recording studio time for five minutes of narration (UN-style voice-over) and even more if you want professional lip-sync dubbing. The new audio is recorded by a professional voice actor who not only is fluent in the foreign language but can deliver it with expert rhythm and timing to match the original audio as close as possible. Once the new audio is recorded, it then takes time to clean up the voice track and equalize audio levels, etc.
Typically this is done when someone is being interviewed with the assistance of an interpreter. You have probably seen this when watching a news reporter interview an athlete or politician who speaks a foreign language. The speaker will begin speaking in his or her native tongue. After a few seconds, their audio drops slightly and the interpreter will begin narrating. By allowing the original native language to still be audible, it preserves the authenticity, tone, and dramatic effect of the character.
You have probably seen this when watching old kung fu movies where the original Chinese dialogue has been replaced with English or any other language. You cannot hear the original audio track and the new voice-over is made to look as if the actor was originally speaking that language.
Approaches to translation will differ depending on the type of program (for example, if the speaker is on-camera or off-camera). A direct verbatim translation is not necessarily the best as it typically becomes expanded and lengthy. Instead, an adaptation of the original narration is preferred. If lip-sync dubbing or looping is desired, the translation may be even more time-consuming as syllable counts must be the same in both languages and words beginning with letters that denote specific mouth movements (like O, A, M, P, S, T, R) are matched as much as possible.