A recording contains a crash, followed by a speaker saying, “That explains it.” A words-only transcript preserves the reply but may remove the event that makes it understandable.
When editing a transcript, identify sounds that affect meaning: an alarm discussed by the speaker, applause that interrupts an answer, or laughter that changes the tone. Avoid cataloguing every incidental movement simply because you can hear it.
The W3C transcript guidance is a useful reference for accessible text alternatives. It also explains that clarifications added to a transcript should be distinguishable from recorded content.
Describe what the recording supports
“[Glass breaks]” may be more specific than the evidence permits if the source is unseen and uncertain. “[Sharp crash]” records the audible event without supplying an object you cannot identify.
Keep the placement close to the relevant speech. Putting all sound descriptions at the end leaves readers to reconstruct the sequence themselves.
After adding a label, read the sentence on either side without playing the audio. Does the label restore an important relationship, or interrupt the conversation with unnecessary detail? A second review focused on this question is more useful than making every bracket equally elaborate.
Editorial illustration from this site’s image library; not documentary evidence of the example or object discussed.