Speech balloons are exclusively tied to someone or something uttering them – persons, telephones, loudspeakers, animals, etc. If an utterance has a ‘tail’ pointing towards someone or something, it is almost always a speech balloon. The same is also the case for thought balloons.
A caption is detached from a speaker. Captions are usually boxes indicating a change of place and time (“Meanwhile,” “And Then,” etc.), but they can also contain narration. Captions do not have a tail.
Onomatopoeia are ‘written sounds’, like “screech,” “boom,” “honk,” and so on.